Outdoor

Seamless Pedestrian Navigation in Indoor/Outdoor Large Spaces with No Clear Patterns for Movement PhD Research Proposal

Jinjin YAN, MSc PhD candidate 2016 - 2020

January 18, 2018

Faculty of Architecture and the Built Environment

iii

Contents 1

Introduction 1.1 Motivation . . . . . . . . . . 1.2 Expected Research Outcome 1.3 Scope of the Research . . . . 1.4 Structure of this report . . .

. . . .

1 1 4 4 5

2

Literature Study 2.1 Spaces for Pedestrian Navigation . . . . . . . . . . . . . . . . . . . . . . 2.1.1 Indoor & Outdoor Space Definition . . . . . . . . . . . . . . . . 2.1.2 Semi-indoor Space Definition & Semi-outdoor Space Definition 2.1.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Space Subdivision based Pedestrian Navigation Models Construction . 2.2.1 Poincaré Duality for Navigation Models Construction . . . . . 2.2.2 2D Approaches for Navigation Models Construction . . . . . . 2.2.3 3D Approaches for Navigation Models Construction . . . . . . 2.2.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3 Contexts in Space Subdivision . . . . . . . . . . . . . . . . . . . . . . . .

7 7 7 8 10 11 12 13 17 19 20

3

Proposed PhD Research 3.1 Problems in Pedestrian Navigation . . . . . . . 3.2 Research Questions . . . . . . . . . . . . . . . . 3.3 Methodology . . . . . . . . . . . . . . . . . . . 3.3.1 Spaces . . . . . . . . . . . . . . . . . . . 3.3.2 Spaces-subdivision . . . . . . . . . . . . 3.3.3 Context Related to Space Subdivision. . 3.3.4 Implementations & Case Study Models 3.4 Preliminary Results . . . . . . . . . . . . . . . . 3.4.1 Pedestrian definition . . . . . . . . . . . 3.4.2 Spaces Classifications and Definitions . 3.4.3 Space Management . . . . . . . . . . . . 3.4.4 Experiments of Enclosing Spaces . . . .

23 23 24 25 25 27 30 32 34 34 35 39 41

4

Practical Issues 45 4.1 Courses & Supervision . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45 4.2 Software, Tools and Data . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

5

Deliverables 5.1 Outcomes and Reports . 5.2 Time Planning . . . . . . 5.3 Short term first year plan 5.4 Publications . . . . . . .

Reference

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . . . . . . . . . .

. . . .

. . . .

. . . . . . . . . . . .

. . . .

. . . .

. . . . . . . . . . . .

. . . .

. . . .

. . . . . . . . . . . .

. . . .

. . . .

. . . . . . . . . . . .

. . . .

. . . .

. . . . . . . . . . . .

. . . .

. . . .

. . . . . . . . . . . .

. . . .

. . . .

. . . . . . . . . . . .

. . . .

. . . .

. . . . . . . . . . . .

. . . .

. . . .

. . . . . . . . . . . .

. . . .

. . . .

. . . . . . . . . . . .

. . . .

. . . .

. . . . . . . . . . . .

. . . .

. . . .

. . . . . . . . . . . .

. . . .

. . . . . . . . . . . .

. . . .

47 47 47 48 48 49

v

List of Figures 1.1

Examples of current problems in pedestrian navigation. . . . . . . . . .

2.1 2.2

Gas station. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 Space is classified and defined into four types according to the number of available satellites for positioning (Wang et al., 2016). . . . . . . 10 Three indoor/outdoor environment types and the representative scenes (Zhou et al., 2012). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 Poincaré duality (Li, 2016). . . . . . . . . . . . . . . . . . . . . . . . . . . 12 Grid tessellation of indoor space (Afyouni, Cyril, and Christophe, 2012). 13 Navigation network derived using MAT (Kallmann and Kapadia, 2014). 14 Space partition according to Visibility Graph (Stoffel, Lorenz, and Ohlbach, 2007). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 A set of polygonal objects (a), the associated visibility graph (b), and a shortest path (c) (Berg et al., 2013). . . . . . . . . . . . . . . . . . . . . 15 Adaptive extended Voronoi diagram (Hilsenbeck et al., 2014). . . . . . 16 Space partition based on irregular triangulations: CDT approach (Borovikov, 2011) and Constrained TIN (Xu, Wei, and Zlatanova, 2016). . . . . . . . 16 Representing and subdividing space by using cubes (voxels) (Yuan and Schneider, 2010) and Octree (Rodenberg, Verbree, and Zlatanova, 2016). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 Example 3D volume based complete space subdivision in an indoor building (Boguslawski and Gold, 2009). . . . . . . . . . . . . . . . . . . 18 Space subdivision and navigation network based on convex polyhedron (Diakité and Zlatanova, 2017). . . . . . . . . . . . . . . . . . . . . . 19

2.3 2.4 2.5 2.6 2.7 2.8 2.9 2.10 2.11

2.12 2.13 3.1 3.2 3.3 3.4 3.5 3.6 3.7 3.8 3.9 3.10

3.11 3.12 3.13

Walls in volumetric-based space model. . . . . . . . . . . . . . . . . . . Wallsurface/‘paper’ wall in the CityGML (Gröger et al., 2008). . . . . . Example of the distance between people and wall in a straight corridor (Bosina et al., 2016). . . . . . . . . . . . . . . . . . . . . . . . . . . . Possible subdivision approach for indoor space (Diakité and Zlatanova, 2017). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Outdoor navigation network based on space subdivision (Rusné, Ljiljana, and Beirao, 2017). . . . . . . . . . . . . . . . . . . . . . . . . . . . . Closing spaces for space subdivision. . . . . . . . . . . . . . . . . . . . . Patterns of space visiting. . . . . . . . . . . . . . . . . . . . . . . . . . . Information desk with functional areas (coloured areas). . . . . . . . . Workflow of my research. . . . . . . . . . . . . . . . . . . . . . . . . . . (a) is the 2D map of the imaginary small city; (b) is the 3D map of the navigation area; (c) is the spaces is pedestrians can use; (d) is the sI-spaces and sO-spaces. . . . . . . . . . . . . . . . . . . . . . . . . . . . Planned experimental scenarios in TU Delft. . . . . . . . . . . . . . . . Proposed definition and types of the pedestrians. . . . . . . . . . . . . Features of the spaces. . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2

26 27 27 28 28 29 30 32 32

33 33 34 35

vi 3.14 3.15 3.16 3.17 3.18 3.19 3.20

Example of the roof, top, wall and side. . . . . . . . . . . . . . . . . . The workflow of proposed spaces definition framework. . . . . . . . Definitions of the spaces based on proposed definition framework. . Bridges between two buildings. . . . . . . . . . . . . . . . . . . . . . . Some examples of the spaces. . . . . . . . . . . . . . . . . . . . . . . . The volumetric-based data model of spaces. . . . . . . . . . . . . . . . The basic forms of roofs and roof projections (Neufert, Neufert, and Kister, 2012). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.21 Three examples of space creating. . . . . . . . . . . . . . . . . . . . . . 3.22 Inner walls cut the floor space into room spaces. . . . . . . . . . . . . 5.1

. . . . . .

36 37 38 39 40 41

. 42 . 42 . 43

Time planning overview. . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

vii

List of Tables 2.1 2.2

Existing approaches for navigation models construction. . . . . . . . . 12 Examples of criteria for determining the functional space of the ob¯ jects in indoor spaces subdivision (Kruminait e˙ and Zlatanova, 2014). . 21

3.1 3.2

Literature review of pedestrian preferences for the spaces. . . . . . . . 25 ¯ ˙ Criteria and value ranges of functional spaces in indoor space (Kruminait e, 2014). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

4.1

The courses in the first year. . . . . . . . . . . . . . . . . . . . . . . . . . 45

5.1

Key events in the first year of research. . . . . . . . . . . . . . . . . . . . 48

ix

List of Abbreviations MAT CDT NRS NRG MLSM GPS LBS RFID TIN DG VG GNM 1D 2D 3D IPS INS OGC POI IFC BIM

Medial Axis Transform Constrained Delaunay Triangulations Node-Relation Structure Node-Relation Graph Multi-Layered Space Model Global Positioning System Location Based Services Radio Frequency Identification Triangulated Irregular Network Dual Graph Visibility Graph Geometric Network Model One Dimensional Two Dimensional Three Dimensional Indoor Positioning Systems Inertial Navigation System the Open Geospatial Consortium Point of Interest Industry Foundation Classes Building Information Modelling

1

Chapter 1

Introduction 1.1

Motivation

Navigation, also called path-finding or way-finding, whether in the real or electronic world, is a fundamental activity. The navigation is described as the method of determining the direction of a familiar goal across unfamiliar terrain (May et al., 2003), or the process of orientation to reach a specific distant destination from the ¯ origin (Kruminait e˙ and Zlatanova, 2014). Seamless pedestrian navigation (universal/continuous navigation in both indoor space and outdoor space) is very important to us, since, during the navigation process, indoor space and outdoor space do not exist in isolation and pedestrians can move seamlessly in both types of spaces (Nagel et al., 2010). To be able to perform correct navigation several components have to be available (Zlatanova et al., 2014; Worboys, 2011): (3D) localisation of start point and destination, a (3D) model that represents the space subdivision, (3D) algorithms for path computation (on a topological model or a grid), guidance (Points of Interest, Landmarks), guidance (visualisation) of the path and finally tracking/correction (if the path is not followed). In the past decade, various integrated navigation systems have been developed to provide seamless pedestrian navigation. Examples of the systems are, GPS positioning plus Inertial Navigation System (INS) in (Cheng et al., 2014), combining GPS (for outdoor positioning) and QR codes (for indoor positioning) (Nikander et al., 2013; Shelke et al., 2016), the system of (Kourogi et al., 2006) based on GPS and active Radio Frequency Identification (RFID) tag, etc. Nevertheless, those systems are mainly focused on localisation of the start point and destination, rather than the whole process of seamless pedestrian navigation. Furthermore, in these systems, navigation solutions for pedestrians are separated in two worlds, indoor world (space) and outdoor world (space). That is, the indoor and outdoor spaces are treated separately, instead of the same (unified) way, e.g., the same structure, the same management methods, the same space subdivision rules, navigation model with the same structure, and using the spaces in the same way. Therefore, the current seamless navigation is still not very successful. Space is crucial for pedestrian navigation, because the latter can be regarded as the movements of pedestrian from one space to another connected space. The essence of space, from the point of view of pedestrian, is the same in both indoor and outdoor environments (Yang and Worboys, 2011a): pedestrians are only able to walk freely within these spaces if they are not occupied by objects (e.g. furniture), or are not restricted by either implicit or explicit boundaries (e.g. areas dedicated to vehicles or staff members, for the former, or fences, etc., for the latter). Pedestrians have the same patterns of movement in both indoor and outdoor, i.e., they try to avoid obstacles or select the most convenient path for their purpose in the same way. This implies, that, both for indoor and outdoor, it is necessary to identify the

2

Chapter 1. Introduction

E Door2 Path 1

Door1

Path 2

Path 3 S

(a) Network glued by anchor node

(b) Limited seamless navigation

E

(c) Guide pedestrians to walk in the street

(d) Unrealistic detour in the square

F IGURE 1.1: Examples of current problems in pedestrian navigation.

accessible spaces for the purpose of pedestrian navigation. Pedestrians can have well-defined preferences on how to pass through space and those preferences may depend on multiple factors, e.g. time of the day, weather conditions, emergency situations or even gender differences (Melson, 1977). However, until now outdoor pedestrian navigation does not take space into account, so pedestrians fail to find the information of specific spaces for user-adaptive paths. Thus, current solutions of outdoor pedestrian navigation are not sufficient, making the seamless navigation unreachable. Further, shortcomings of in current indoor and outdoor seamless pedestrian navigation can be summarised as follows: • A lot of navigation models subdivide the spaces in different ways to fulfil their dedicated applications, leading to these models cannot be combined directly. There are many indoor navigation models construction methods based on space subdivision, in which the space units (sub-spaces) could be functional spaces ¯ (Kruminait e˙ and Zlatanova, 2014; Diakité and Zlatanova, 2017), rooms (Teo and Cho, 2016), even entrances (or doors) (Zlatanova et al., 2014) in different applications. Therefore, these models cannot be integrated into a unified one. Even if they are glued together compulsively, the “unified” one still cannot meet the navigation needs. For instance, a pedestrian will fail to find a printer in a very big/wide room, if the space units of the navigation model represent rooms only, rather than functional spaces in rooms. On the contrary, if this pedestrian only wants to pass through this building with the shortest path, but the space units are functional spaces, and too much unnecessary computing resources will be involved in paths computation. • The current outdoor pedestrian navigation methods are relying on systems initially not designed for humans.

1.1. Motivation

3

Current outdoor pedestrian navigation is based solely on a specific navigation network which originally designed for vehicle navigation, and does not take space into account. The consistent practice for pedestrian navigation in outdoor space obtains the pedestrian navigation network by reusing and adapting the road networks based on pedestrian characteristics, e.g., speed (Corbetta et al., 2016; Ziemer, Seyfried, and Schadschneider, 2016), dimension (Liu and Zlatanova, 2015), flexibility (Gaisbauer and Frank, 2008), behaviours (Hughes, 2002; Hoogendoorn and Bovy, 2004; Bosina et al., 2016; Köster, Lehmberg, and Dietrich, 2016; Daamen, Bovy, and Hoogendoorn, 2002), etc. Then, this outdoor navigation network is glued to indoor navigation network by an anchor, see Figure 1.1(a). In reality, compared to vehicles, pedestrians have more freedom in outdoors, but current pedestrian navigation models are limited because they failed to represent somewhere. For example, pedestrians can cross a road lane at any desired location regardless whether there is a crosswalk present, but these desired location are failed to be modelled in current navigation network. In addition, some outdoor spaces are inaccessible for pedestrians, e.g., high ways or bicycle lanes with fences. The navigation paths given by Google Maps in Figure 1.1(d) from the start point (S) to the destination (E) only based on the road network, so they navigate pedestrians to make a detour around the Delft square. But the reality is that pedestrians can walk through the square directly. Therefore, the solution of pedestrian navigation in outdoor only by reusing and adapting the road networks, but without considering the space, is not sufficient for outdoor pedestrian navigation. • The seamless pedestrian navigation is limited.

Because a variety of buildings have no existing navigation models, the existing seamless navigation tends to offer outdoor navigation paths as much as possible, rather than trade-off paths with both indoor and outdoor. Moreover, pedestrians do not just want to get a general navigation path, but a path that meets their specific (user-adaptive) requirements. For instance, if a pedestrian forgets to take an umbrella, or sudden change of the weather occurs, in a rainy day, he/she may prefer to walk as much inside of buildings (or somewhere with shelters) as possible way to escape from the rain. More examples are that large road with side-walks, which are safer for pedestrians, may be preferable to the shortest route (Koide and Kato, 2005). Figure 1.1(b) illustrates navigation paths for pedestrians from the start point (S) to the destination (E), in which the Google Maps showed three paths. All the paths are meant to guide pedestrians out of the building (indoor) as soon as possible to reach the outside (outdoors). Paths 1 is a more suitable path than Path 2 and 3 in rainy day for the pedestrians who have no umbrellas, because it includes part of indoor path. However, pedestrians prefer to walk inside the building and go outside by the Door 1 or Door 2 to reduce the exposure in outdoor. Another situation is that the navigable space for pedestrians is unsuited. For instance, the seamless navigation path guides pedestrians to walk in the street, which is not allowed in real life considering the safety of pedestrians, see Figure 1.1(c). These examples show that the current seamless indoor and outdoor pedestrian navigation is still very limited.

In summary, current solutions for seamless pedestrian navigation is not enough, since (a) different spatial models are dedicated for different applications, so these

4

Chapter 1. Introduction

models are incompatible to construct a unified one; (b) the current outdoor pedestrian navigation does not take space into account, thus pedestrians fail to find the information of specific spaces for user-adaptive paths; (c) fewer research is focused on the issue that to share the same subdivision rules from indoor to outdoor pedestrian navigation network construction, as well as same division rules for different applications. Considering indoor navigation network for pedestrians has been more investigated, and the essence of the space in indoor and outdoor is the same from the perspective of pedestrian navigation, it is worth to investigate how to build a framework to construct a unified 3D spatial model and share indoor space subdivision methods (e.g., 3D space-subdivision plus Poincaré duality) with outdoor space to construct a unified navigation network for seamless navigation.

1.2

Expected Research Outcome

The goal of this research is to develop: A framework to construct a unified 3D spatial model for space-subdivision-based pedestrian navigation in indoor and outdoor, that can take into account specific contexts. To achieve this goal, this research will include defining relevant spaces (indoor, outdoor, semi-indoor, and semi-outdoor), space management, determining relevant objects based on the pedestrian contexts in subdivision, 3D space subdivision based on convex polyhedron (cells with variable size), navigation network construction for the spaces, and paths computation (for tests).

1.3

Scope of the Research

• The research will focus on space for seamless pedestrian navigation, including a unified space model (indoor and outdoor), and 3D space subdivision methods for seamless navigation model construction. The navigation paths will be computed in some cases at the end of this research to test the unified space, space subdivision approaches and navigation network construction. • In this research, the navigation agent is pedestrian, and not vehicles. The definition of the pedestrian can be seen in the Chapter 3. But I will not do research on pedestrian itself, for instance, operational (walking), tactical (short term path choice) and strategical (route choice, activity chaining) behaviours. Instead, I will provide a solution to embed these aspects into navigation. Specifically, I will take some existing research outputs of the pedestrian as inputs (references) to decide what kind of spaces should be considered in pedestrian navigation in my research. For instance, the space under the overhanging roof will be considered as some pedestrians are interested in it in a rainy day, but I will not investigate why and who. • Contexts will be involved, but only limited to determine which objects will be included in the space subdivision. For instance, tourists may be interested in the best camera points, benches, and open space can be used for picnic, but these spaces are not interesting for pedestrians who want to walk with the dog for leisure only. Thus, the purposes and types of the pedestrians are contexts in this research. Based on the contexts, we can determine that the best camera points, benches, and open space can be used for a picnic will be

1.4. Structure of this report

5

included in the space subdivision for tourists, but not in the space subdivision for the pedestrians who walk with dogs for leisure. I will not study these contexts themselves. Instead, I will provide a solution that can offer different navigation options for different contexts. • This research only takes the CityGML1 , IndoorGML2 , and IFC/BIM3 as the data sources, and tailors them to fit the space computations, because these are the related prevailing international standards. Furthermore, the 3D indoor and outdoor geometry models construction approaches are not included in this research, but I will build some 3D geometry models manually to prove the concepts that I will develop. • This research will not touch the positioning related topics, for example, GPS, sensors, positioning theories, methods, accuracy, and algorithms, etc. Moreover, seamless indoor and outdoor navigation based on the system integration (e.g., integration of GPS and INS) is not included.

1.4

Structure of this report

This document comprises five chapters. • The current chapter introduces the motivation, the research outcome, and the scope of the research. The remaining chapters are as follows: • Chapter 2 summaries background for this research. Several essential concepts of indoor and outdoor navigation are introduced, including spaces, spacesubdivision-based navigation network, and the contexts in pedestrian navigation. • Chapter 3 is used to describe problems in current pedestrian navigation, research questions, methodology, and some preliminary results, in which we refine one main research question with three sub-questions. • Chapter 4 involves some specific practical issues, including courses, supervision, software, tools and data. • Chapter 5 indicates the proposed deliverables of this research, and also the time planning.

1

https://www.citygml.org http://indoorgml.net 3 http://www.buildingsmart-tech.org 2

7

Chapter 2

Literature Study There are a variety of research focused on indoor and outdoor pedestrian navigation, including spaces and space modelling (Becker, Nagel, and Kolbe, 2009a; Lee et al., ¯ 2014; Boguslawski and Gold, 2009), space subdivision (Kruminait e˙ and Zlatanova, 2014; Zlatanova et al., 2014; Diakité and Zlatanova, 2017), navigation network models (Liu and Zlatanova, 2011; Mandloi and Thill, 2010; Teo and Cho, 2016), context in navigation (Afyouni, Ray, and Claramunt, 2010; Afyouni, Cyril, and Christophe, 2012; Becker, Nagel, and Kolbe, 2009b; Khan, Yao, and Kolbe, 2015; Saeedi, 2013; Wu et al., 2007), and indoor and outdoor seamless navigation (Basiri et al., 2016; Cheng et al., 2014). In this chapter, considering the expected outcome of this research (a 3D spatial model for space-subdivision-based pedestrian navigation in indoor and outdoor, that can take into account specific contexts), I will focus on three main points in current research: spaces for pedestrian navigation, subdivision of spaces, and the specific context.

2.1

Spaces for Pedestrian Navigation

A clear classification, and definition of spaces for pedestrian navigation is the basis for all subsequent studies, e.g., space management, 3D spatial model design, and space subdivision. Therefore, in this section, works related to where classification and definition of spaces from different sources are presented. In current research, space for walking is classified into two categories, indoor space, and outdoor space. Indoor space, can be further classified into two types (Slingsby and Raper, 2008): navigable space, and non-navigable space, or into three types (Zlatanova, Liu, and Sithole, 2013; Diakité and Zlatanova, 2017): object space, functional space, and free space.

2.1.1

Indoor & Outdoor Space Definition

Indoor and outdoor spaces possess different characteristics, such as constraints (Li, 2008; Yan et al., 2016), scale and dimension (Yang and Worboys, 2011b; Richter, Winter, and Santosa, 2011). They are defined in many dictionaries and contemporary research. Indoor and outdoor space definition can be found in many dictionaries, such as the Oxford Dictionary, Merriam-Webster, Dictionary, Cambridge Dictionary, Collins Dictionary, TheFreeDictionary. The definitions of indoor space from dictionaries relates to the physically enclosed space, such like the space inside or the interior of a building or a house. While the space around the building or house, out of doors, in the open air, or wildness is outdoor.

8

Chapter 2. Literature Study

In addition to the definitions from dictionaries, indoor and outdoor specifications can also be found in a lot of contemporary research. In Indoor Geography Markup Language (IndoorGML) (Lee et al., 2014), indoor space is all space within one or multiple buildings consisting of architectural components. Afyouni, Ray, and Claramunt (2010) defined indoor space as a building environment (such as a house, a commercial shopping centre, etc.), which people usually behave in. Indoor space in (Yang and Worboys, 2011a) refers to built rather than natural environments. It covers the enclosed interiors of buildings above the ground and spaces underneath the ground that afford platforms for human activities. Winter (2012) defined the indoor space with analogy to the human body. The body is a container enclosed by skin. Because of the skin, the body becomes isolated from the outside, so the skin brings the concept of inside and outside to the body. Similarly, the wall, floor, roof, fence, are skin, which brings the concept of inside and outside to space. In this metaphor, inside of the skin is indoor space, outside the skin is outdoor space and the doors are entrances. Indoor space in (Zlatanova et al., 2014) is an artificial construct designed, and developed to support human activities. Virtual representations of indoor space have to be able to support these activities. Unlike for indoor spaces, there is no formal definition of the outdoor space, but we can take the outdoor space as the space out of indoor space.

2.1.2

Semi-indoor Space Definition & Semi-outdoor Space Definition

It is insufficient to classify and define the space for walking into indoor and outdoor, since pedestrians inevitably need to move into intermediary spaces between indoors and outdoors, or semi-enclosed spaces (e.g. the space under an overpass, bus shelters, porch, etc.). Such types of spaces can play an important role in navigation options for pedestrians, but they cannot be categorised as indoor (not inside a building), nor outdoor (not in the open air), e.g., the gas station, see Figure 2.1. These spaces may have characteristics of both indoor and outdoor spaces. Often such spaces act as connections between indoor and outdoor environments, an approach which has been used to improve residential comfort and reduce cooling and heating energy requirements (Kim, Kim, and Leigh, 2011), as well as a special structure to obtain a good building micro-climate (Du, Bokel, and Dobbelsteen, 2014). Moreover, the micro-climate of the "semi-outdoor" (partially enclosed space) often has a better thermal sensation (usually lower effect of wind, less heat) than the outdoor (Pagliarini and Rainieri, 2011b). Therefore, people may prefer to use them for some additional purposes, e.g., drying laundry, growing plants (Kim, Kim, and Leigh, 2011), learning activities (Nasir, Salim, and Yaman, 2014), adapting to thermal environments (Lin, 2009), sheltering from the sun and wind (He and Hoyano, 2010), and improving the physical environment of the markets(Kim, Park, and Kim, 2008). In the user-adaptive pedestrian navigation, pedestrians maybe prefer to visit these spaces for some purpose (e.g., escape from rain). However, while such semi-spaces have been investigated in other domains (e.g., architectural design for residential comfort improvement, spatial design for good building micro-climate obtaining), they are inexistent in the researches related to navigation. The definition of semi-indoor space is often related to the presence of roof, because it plays a key role in contributing to the climate control. Turrin et al. (2009) defined the roof-covered space as semi-indoor space, which is partially surrounded by adjacent indoor space. Van Timmeren and Turrin (2009) covered outdoor space by the Vela Roof to create semi-indoor space to passively avoid uncomfortable (cold or

2.1. Spaces for Pedestrian Navigation

9

F IGURE 2.1: Gas station.

overheated) conditions, thereby further reducing energy demand for thermal comfort. Hooff and Blocken (2009) clearly created and defined the semi-indoor space in the research titled Computational Analysis of Natural Ventilation in a Large Semi-enclosed Stadium. Specifically, by closing the roof, a semi-indoor environment is created and spectators and equipment are protected from wind, rain, and snow. The example is a semi-indoor stadium, which is characterized as a stadium that has a roof that can be used to close the indoor volume to a relatively large extent, but that even in this setting still has direct openings to the outside. However, Bouyer et al. (2007) took the stadium as semi-outdoor space in the assessment research of thermal comfort. Indoor and semi-indoor are also defined as GPS-denied scenarios in (Ortiz et al., 2015). The definition of semi-outdoor space is also given in many works of literature, especially from the assessment of thermal comforts. Pagliarini and Rainieri (2011a) took the space which is partially open towards the outdoor environment as semioutdoor space in the research of dynamic thermal simulation of a glass-covered semi-outdoor space with roof evaporative cooling. Similarly, they reinforce the concept that space enclosed by a semi-transparent pitched roof is semi-outdoor space. In a later thermal environment characterization research (Pagliarini and Rainieri, 2011b), a glass roofing outdoor space was subjected as semi-outdoor space. Spagnolo and De Dear (2003) defined the semi-outdoor as locations that, “while still being exposed to the outdoor environment in most respects, include man-made structures that moderate the effects of the outdoor conditions.” Examples include roofs acting as radiation shields or walls acting as vertical windbreaks. Lin, Matzarakis, and Huang (2006) took bus shelters as an example of semi-outdoor space, as they offer shelter provided in the form of a roof. The semi-outdoor space in (Yang, Wong, and Jusuf, 2013) refers to the internal architectural space in a given building with maximum exposure to the outdoor climate such as lobbies, corridors, atrium, courtyards, passages, and verandas. Goshayeshi et al. (2013) defined the semi-outdoor spaces as the spaces that are partly open in the direction of the outdoor circumstance. Hwang and Lin (2007) defined the semi-outdoors as ‘exterior spaces that are sheltered and attached to the building’. A specific group of semi-outdoor space in (Turrin et al., 2012) can be identified as covered by large roofs or partially open envelopes, leaving a relevant direct connection with the outdoor environment. In addition to the previously introduced architectural definitions, space can be classified according to the number of available satellites for positioning (Figure 2.2). Open Outdoors is the area, which has open sky condition, and people can have access to sufficient satellites for positioning. Semi-outdoors is the GNSS-hostile outdoor environment and without enough satellites for positioning, such as an urban

10


F IGURE 2.2: Space is classified and defined into four types according to the number of available satellites for positioning (Wang et al., 2016).

F IGURE 2.3: Three indoor/outdoor environment types and the representative scenes (Zhou et al., 2012).

canyon or a wooded area. Light indoors is similar to semi-outdoors as it has few satellites available due to the windows or other openings, while the Deep indoors environment refers to a place without any available satellites at all. Another definition is from (Zhou et al., 2012), which considers buildings as the main objects. They classified the environment into three categories (i.e. outdoor, semi-outdoor, and indoor) to provide finely-grained context information for upper-layer applications. To be specific, the space outside of a building is outdoor, while inside a building is indoor. Close to or semi-open building space is semi-outdoor. Figure 2.3 illustrates the definitions, examples and representative scenes for those three different spaces. The two definition methods based on different criteria and applications are the same in essence, i.e., the space is categorized into three types, indoor, outdoor, and semi-outdoor. However, there are some scenes that do not fit within above definitions, for example, the gas station, see Figure 2.1. The gas station is not an indoor space, because it is not bounded completely; nor an outdoor space, because of the roof; nor a semi-outdoor space, because it is not near a building while it is possible to get enough satellites for positioning; nor a light indoor, because it is not a room with windows and lastly deep indoor is also not appropriate.

2.1.3

Summary

In summary, indoor space and outdoor space are well defined to a certain degree, but the intermediary spaces/semi-enclosed spaces are ill-defined, such as balconies, overhanging roofs, backyard, spaces under bridges and so forth. Literature defines

2.2. Space Subdivision based Pedestrian Navigation Models Construction

11

them as semi-indoor space or semi-outdoor space from multiple perspectives and in different ways. The reason why they defined the space as semi-indoor space is that these spaces have the roof but they are also partly open to the outdoor environments. Additionally, for the same reason, other research has defined them as semi-outdoor spaces. Therefore, the current definition of the ill-defined spaces has contradictions. For instance, a stadium space, because of the roof, is regarded as semi-indoor space in (Hooff and Blocken, 2009), but with the same reason, it is defined as semi-outdoor space in (Bouyer et al., 2007). The definitions in (Wang et al., 2016; Zhou et al., 2012), to some extent, are reasonable, but they do not include some semi-enclosed spaces, e.g., the gas station seen in Figure 2.1. The definition varies from person to person, from application to application, even for the same space, e.g., the stadium. In current research, the ill-defined spaces can be regarded as one of the four: - Outdoor (Van Timmeren and Turrin, 2009). The space that is not completely enclosed by walls, windows and doors may be considered as outdoor space; - Semi-indoor (Turrin et al., 2009; Kim, Kim, and Leigh, 2011; Chengappa et al., 2007; Tyson, Schuler, and Leonhardt, 2012; Amutha and Nanmaran, 2014; Liu et al., 2013; Kim, Park, and Kim, 2008; Monteiro and Alucci, 2007). Space covered by canopies is related to the buildings, can combine indoor and outdoor climate and therefore is considered as semi-indoor; - Semi-outdoor (Du, Bokel, and Dobbelsteen, 2014; Hwang and Lin, 2007; Lin et al., 2008; Indraganti, 2010; Pagliarini and Rainieri, 2011b; Cao et al., 2017), semi-enclosed space (He and Hoyano, 2010; Kim, Kato, and Murakami, 2001), or semi-open space (Philokyprou et al., 2017). Space that is not enclosed although has man-made structures that moderate the effects of the outdoor conditions can be regarded as semi-outdoor; - Connection/transition/buffer areas (Slingsby and Raper, 2008; Li, 1994). Therefore, a comprehensive space-definition framework to define all of the spaces in a uniform way is needed, so as to improve the seamless and user-adaptive pedestrian navigation.

2.2

Space Subdivision based Pedestrian Navigation Models Construction

Currently, there are a variety of approaches for indoor pedestrian navigation models construction, but less for outdoor space. In other words, indoor navigation models for pedestrians have been more investigated than outdoor. Largely only geometry of the spaces is involved in current indoor navigation models construction, and existing methods include 2D and 3D approaches (Zlatanova et al., 2014). In general, grid, and network the two most used kinds of navigation models. Although these methods have different strategies, as well as different criteria, they have the same goal to construct the pedestrian navigation network. Examples can be seen in Table 2.1. More details about the approaches for navigation models construction can be seen in the following subsections.

12

Chapter 2. Literature Study TABLE 2.1: Existing approaches for navigation models construction. 2D approaches

3D approaches

Grid models

Square-shaped grid (Afyouni, Cyril, and Christophe, 2012), Hexagonal grid, etc.

Voxels (Yuan and Schneider, 2010; Nourian et al., 2016), Octree (Fichtner, 2016; Rodenberg, Verbree, and Zlatanova, 2016), etc.

Network navigation models

MAT (Kallmann and Kapadia, 2014), VG (Stoffel, Lorenz, and Ohlbach, 2007), Voronoi diagrams (Hilsenbeck et al., 2014), CDT (Borovikov, 2011) or constrained TIN (Xu, Wei, and Zlatanova, 2016)

Convex polyhedron (Diakité and Zlatanova, 2017), etc.

2.2.1

Poincaré Duality for Navigation Models Construction Primal Space

from

to

Dual Space

Primal Space

from

to

2 dim. 0 dim.

3 dim. 0 dim.

1 dim. 1 dim.

2 dim. 1 dim.

1 dim. 1 dim.

1 dim. 2 dim.

0 dim. 2 dim.

0 dim. 3 dim.

(a) 2D

(b) 3D

Dual Space

F IGURE 2.4: Poincaré duality (Li, 2016).

Network navigation model is an explicit model of Node-Relation Graph (NRG) for the pedestrian navigation. Poincaré Duality (Munkres, 1984) provides a theoretical background for network navigation model from space, i.e., theory of mapping spaces to NRG. Specifically, it is the theoretical background to simplify the complex spatial relationships between 3D objects by a combinatorial topological network model. A k-dimensional object in N-dimensional primal space is mapped to a (N-k)-dimensional object in dual space. Solid 3D objects in primal space (e.g., rooms within a building) are mapped to vertices (0D) in dual space. The common 2D face shared by two solid objects is transformed into an edge (1D) linking two vertices in dual space. Thus, edges of the dual graph represent adjacency and connectivity relationships which may correspond to doors, windows, or hatches between rooms in primal space. Figure 2.4 illustrates this duality transformation, in which the size of the minimum/unit navigation space can be critical for navigation network construction. For instance, a room can be treated as a unit space in a floor. But this may not be optimal in some conditions (e.g. in the case of very wide rooms, halls, etc.). Therefore, the antecedent condition of using the Poincaré Duality to construct the appropriate navigation network is proper space subdivision. After the space subdivision, based on the Poincaré Duality (Becker, Nagel, and Kolbe, 2009a), sub-spaces


13

(a) Square shaped grid

(b) Hexagonal grid

F IGURE 2.5: Grid tessellation of indoor space (Afyouni, Cyril, and Christophe, 2012).

can be represented as navigation nodes, and the topological connectivity between spaces are used as navigation nodes. Then, Network navigation model can be constructed.

2.2.2

2D Approaches for Navigation Models Construction

In general, 2D approaches can be categorised into two types: (a) Grid models, such as square-shaped grid, hexagonal grid, etc. (b) Network navigation models. Floor plans are common original inputs for 2D approaches for indoor/outdoor navigation network construction. Medial Axis Transform (MAT) and Visibility Graph (VG) are two approaches without space subdivision, i.e., drive a network for path computation directly from floor plan. Voronoi diagrams, Constrained Delaunay Triangulations (CDT), constrained TIN are three approaches based on irregular space tessellations. Grid models Grid tessellation is a well-known subdivision of space. The space is subdivided into a finite number of non-overlapping areas and various types of information are assigned to the cells (Afyouni, Cyril, and Christophe, 2012) to represent navigable and impassable regions in space by associating different cell states, see Figure 2.5. This method includes regular tessellations and irregular tessellations, in which regular tessellations decompose space into cells that have the exact same shape and size, e.g., primarily square- (Figure 2.5a) and hexagonal-(Figure 2.5b) shaped cells (Galˇcík and Opiela, 2016), while the irregular tessellations aim at providing an adaptive decomposition of space that is suitable to exactly represent the complexity of the environment being studied (e.g., to accurately represent obstacles). This approach commonly used in game industry to navigate agents as the agent does not have to

14


F IGURE 2.6: Navigation network derived using MAT (Kallmann and Kapadia, 2014).

follow strict guidelines and can freely move in space. However, the performances of this model in navigation depends on the size of the model and size of the grid cell. Too coarse grid might cause the loss of important information while overly fine grid might disproportionally increase processing time and consume excessive amounts of memory although it provides precise movement in space. Network navigation models • MAT algorithm is a common approach to derive navigation networks, in which no subdivisions are needed (Figure 2.6). This approach entirely relies on the shape of navigable space, but it provides sufficient navigation path for regular spaces with long polygons. Nevertheless, it is not very appropriate for large open spaces, since the generated network might suggest unrealistic navigation paths that are not taken by people (Kallmann and Kapadia, 2014). • Another common approach for navigation network construction is VG. (Stoffel, Lorenz, and Ohlbach, 2007) developed navigation model where the indoor space is partitioned into convex polygons according to the visibility criterion, see Figure 2.7. In contracts to MAT, the VG provides a direct path to the point of interest, and this method does not rely on the shape of the building spaces entirely. Path connects the begin and end point directly, so, space subdivision is also not needed. A VG is a graph of intervisible locations, typically for a set of points and obstacles in the Euclidean plane. Each node in the graph represents a point location, and each edge represents a visible connection between them. That is, if the line segment connecting two locations does not pass through any obstacle, an edge is drawn between them in the graph. It can provide more realistic navigation paths as the length of the paths are optimal, see Figure 2.8. However, because the VG also connects corners of objects within indoor environment, the derived navigation path may suggest for the user of the navigation system to move too close to the walls or corners of objects within indoor space, which is contrary to the walking habits, that people tend to keep a certain distance to wall or corners of objects in a straight corridor (Bosina et al., 2016). Another disadvantage is that, if there are many obstacles with super complex shapes in environments, the visibility graphs are constructed with a large number of nodes and edges. Thus, a great amounts of storage space and


(a)

15

(b)

F IGURE 2.7: Space partition according to Visibility Graph (Stoffel, Lorenz, and Ohlbach, 2007).

(a)

(b)

(c)

F IGURE 2.8: A set of polygonal objects (a), the associated visibility graph (b), and a shortest path (c) (Berg et al., 2013).

time consuming calculations of navigation network and paths might be the result. • Voronoi diagram1 or adaptive extended Voronoi diagram in (Hilsenbeck et al., 2014) is also an excellent choice for partitioning (triangulation) of large space for navigation, see Figure 2.9. The figure is an adaptive extended Voronoi diagram extracted from a floor plan. The left is a two-dimensional floor plan and its skeleton, while the right is the final graph after sampling the skeleton (onedimensionally) and enclosed free space (two-dimensionally). The structure of a building typically not only imposes hard constraints on where a pedestrian can walk (walls). It also outlines important prior where people are likely to walk and where are not allowed to walk. Narrow spaces, for instance, are usually traversed from one end to the other. In contrast to robotic applications, sub-meter accuracy is not required for pedestrian navigation systems, and such space representation is not suitable for accurate guidance. • CDT is one of the most common approaches used to derive navigation network (Figure 2.10a). This approach has two advantages: the network is easy to maintain, because the triangles are one of the simplest features for computers to processes, and it enforces obstacle constraints as specific segments are included in the triangulation process. On the other hand, the CDT allows 1

https://en.wikipedia.org/wiki/Voronoi_diagram

16


F IGURE 2.9: Adaptive extended Voronoi diagram (Hilsenbeck et al., 2014).

(a) CDT approach

(b) Constrained TIN

F IGURE 2.10: Space partition based on irregular triangulations: CDT approach (Borovikov, 2011) and Constrained TIN (Xu, Wei, and Zlatanova, 2016).

determination of several constraints in order to derive the most suitable triangulation for generation of navigation network (Borovikov, 2011). Such constraints might be maximum area of triangle, minimum angle within triangle or maximum number of Steiner points that can be inserted. CDT is suitable for large open spaces as it can provide different movement options. Vertices of triangles, centroids of triangles or centres of triangle edges can be used as nodes in the navigation network. However, navigation network generated using triangulation typically might provide navigation path with unrealistic turns (Afyouni, Cyril, and Christophe, 2012). Constrained TIN is a kind of CDT approach, who takes the holes inside of the space. Therefore, it has similar advantages and disadvantages with CDT. Figure 2.10b shows the space subdivision results by the constrained TIN, in which the D1 and D2 are original point and destination separately. The O1, O2 and O3 are three simplified obstacles with different shapes, and the spaces surrounded by pink lines are subspaces. Then each triangle is represented by a point which becomes a node in the network for navigation. In some cases, a certain feature point (e.g., table, chair, bed, plant), inside the triangle, can be found to represent the area.


(a) Voxels

17

(b) Octree

F IGURE 2.11: Representing and subdividing space by using cubes (voxels) (Yuan and Schneider, 2010) and Octree (Rodenberg, Verbree, and Zlatanova, 2016).

2.2.3

3D Approaches for Navigation Models Construction

The main idea of 3D approaches is keeping the three-dimensional properties of the space, and use the volumes to represent the sub-spaces. The volumes can be different shapes and sizes, such as cubes (voxels), octree nodes, polyhedron (concave or convex), tetrahedron, etc. With the sub-spaces and 3D Poincaré duality theory, navigation network can be derived. The familiar 3D gird methods are Voxels (Yuan and Schneider, 2010) (Nourian et al., 2016), Octree (Fichtner, 2016) (Rodenberg, Verbree, and Zlatanova, 2016). Convex polyhedron (Diakité and Zlatanova, 2017) is one method for network navigation models construction. Grid models • Using cubes (voxels) to represent and subdivide space is one of the most popular in 3D approaches for grid-based models construction. The space is decomposed into cells with same size marked as obstacle or non-obstacle. Based on this representation, routes can be computed by checking the availability of cell movements to their eight neighbours. This model supports navigation in the 3D space by filling out the indoor space with the obstacle and non-obstacle cubes (Figure 2.11a). The obstacle cubes are further classified into insurmountable and surmountable ones to facilitate the 3D navigation. However, the performance of this approach also depends on the size of the unit cubes (voxels). If the voxels are too coarse, important information or space might be lost, while fine voxels can increase the time for processing and need more memory to store, which could be a problematic in the outdoor space. • The octree is also a method used to partition a three-dimensional space by recursively subdividing it into eight octants. Octrees2 are the three-dimensional analog of quadtrees. It is a tree data structure in which each internal node has exactly eight children. Rodenberg, Verbree, and Zlatanova (2016) employed the octree data structure to structure and segment an indoor point cloud for indoor path-finding. An octree consists of a cubical volume which is recursively subdivided into eight congruent disjoint cubes (called octants) until blocks of a uniform colour are obtained, or a predetermined level of decomposition is reached. Compared to the voxels, octree has better performance in the use of 2

https://en.wikipedia.org/wiki/Octree

18


F IGURE 2.12: Example 3D volume based complete space subdivision in an indoor building (Boguslawski and Gold, 2009).

computing resource, since it can represent the point and pointless space by volumes with different size. Moreover, partitioning of space by the octree can result in a hierarchical tree structure, which makes operations of path-finding like neighbour finding. Figure 2.11b shows the octree nodes (Rodenberg, Verbree, and Zlatanova, 2016) to represent the regular indoor sub-spaces, and each space can be a navigation node. • 3D volume-based complete space subdivision is another approach to subdivide the space for navigation network construction. Figure 2.12 illustrates the space subdivision processes and its navigation network, in which S1 and S7 are staircase, S2 S5, S8 S11 are rooms, S6 and S12 are corridor. a) spatial schema, b) volumetric model of rooms, c) complete graph of connections between rooms, d) graph of connections between rooms - only passages. This method can describe the space connections very well, but cannot derive detailed navigation network for accurate pedestrian navigation. Because (a) it abstracts every room as navigation node, rather than subdivides a room space into functional areas. Thus, pedestrians cannot get more detailed information (e.g., printer) in each room; (b) this method assumes that the indoor space is empty (no obstacles), and all the space is navigable space for pedestrians; (c) pedestrian cannot measure the exact distance during the navigation, since the network only contains the connection information. Network navigation models The space subdivision based on polyhedrons is a common method for network navigation model construction, since the polyhedron (sub-spaces) could be limited to concave polyhedron, convex polyhedron, polyhedron with same size, etc. The convex polyhedron is the most popular one as the convex can guarantee the navigation nodes (centroids) located inside of the polyhedrons (sub-spaces) when using


(a)

19

(b)

F IGURE 2.13: Space subdivision and navigation network based on convex polyhedron (Diakité and Zlatanova, 2017).

the Poincaré duality. Figure 2.13 shows an example of space subdivision and navigation network based on convex polyhedron, in which (a) is the sub-spaces of the navigable spaces based on the convex subdivision. Then, the connectivity graph that represents the links between neighbouring subspaces can be derived as shown in (b). The centroids of the convex subspaces are considered as the nodes and edges link of every connected pair of centroids. A quite special approach to 3D space is that 2D navigation network will be embedded in 3D space into 3D navigation network, especially regular buildings (Zlatanova et al., 2014), which is actually not a 3D approach.

2.2.4

Summary

Network navigation model is a way of simulating continuous space with nodes and edges (links). The space is continuous, while the nodes and edges are discrete. Therefore, before using the duality theory to extract nodes and edges from the space, the continuous space should be subdivided. For pedestrian navigation, there are three subdivision levels. We can subdivide spaces into occupied space and unoccupied space. That is the first level of space subdivision. But this level is too general for navigation network construction. So, the second level is, subdividing the unoccupied space into navigable space, and non-navigable space by taking other factors (e.g., pedestrian dimensions) into consideration. Obviously, the occupied space also belongs to non-navigable space. At this moment, navigable spaces are still continuous. To discrete the navigable space, the space will be subdivided (third level) into minimum/unit navigation spaces (subspaces), and each subspace could be a functional space, room, or entrance (or door). Then, navigation network can be constructed based on duality theory, in which navigation nodes are extracted from subspaces, and edges are extracted from topological relationships of subspaces. Compared to grid navigation model, (a) nodes in the network indicate landmarks or decision points and edges between them are the connections. Moving from one node to the other is allowed only when there is an edge between them. Moreover, not only nodes can contain semantic information about the location (name, type, description, etc.), but also other weights can be assigned to the edges (Mortari et al., 2014). (b) assuming human intelligence will compensate for inaccuracies (Zlatanova et al., 2014), network navigation models provide less detailed routing. Thus they enable lower data processing time which is essential in large scenes both

20


¯ ˙ 2014). Therefore, the network-based navindoor and outdoor space (Kruminait e, igation model is more promising, and the most common navigation model used for human navigation. In this research, we will focus on network-based navigation model construction.

2.3

Contexts in Space Subdivision

In order to generate realistic navigation paths for pedestrian, the contexts in space subdivision cannot be ignored. Different kinds of pedestrians have different purposes (in this research, we call them as contexts), and different route choices (Liu and Zlatanova, 2013). Thus, the navigation networks should include different objects (nodes, or sub-spaces) interesting for different kinds of pedestrians. These objects can act as obstacles, destinations, or dynamic factors in the path computations. For instance, the staff pedestrians are interested in the equipments, such as printers, coffee machines, etc., while the visitors like to find the toilets. The similar cases can be found in the outdoor space. For example, the tourists are interested in the attractions, where they can take good pictures, where they can find benches to have a rest, and where they can have a picnic, in which the best camera points, benches, and open space can be used for picnic are interesting objects (spaces) for them, but these objects are not interesting for pedestrians who want to walk with the dog for leisure only. Therefore, the objects should be considered into the navigation are heavily relied on pedestrians, we can analyse them from the types of the pedestrians. According to research Great Streets Initiative3 , there are five reasons why people travelling as pedestrians in both indoor and outdoor spaces, so there are five kinds of pedestrians at least: • Utilitarian Walking - People walk to destinations such as work, school or shopping areas. Most trips include utilitarian walking to reach the final destination. • Rambling - People ramble as a recreational activity, typically for exercise or enjoyment. Rambling may include walking the dog, pushing a baby carriage, jogging, or walking briskly for exercise. • Strolling/Lingering - In certain settings, people stroll and linger. They may stand on the sidewalk and talk with others they meet, sit on a bench, or peoplewatch during an outing. • Promenade - People walk to be seen and interact with other members of the community (e.g. high school students who promenade in groups in shopping malls). • Special Events - People walk at farmer’s markets, public concerts, parades, arts festivals and other community events. In addition to the pedestrian types, the functional spaces of objects also have a significant impact on space subdivision, because the functional areas are not be¯ ing occupied physically, but they are not passable. Functional areas in (Kruminait e˙ and Zlatanova, 2014; Zlatanova, Liu, and Sithole, 2013) are spaces where people are served by a spatial element or are waiting for services provided by the spatial 3

http://www.greatstreetsstlouis.net/residential-neighborhood-design/multimodal-corridorplanning/pedestrians

2.3. Contexts in Space Subdivision

21

TABLE 2.2: Examples of criteria for determining the functional space ¯ of the objects in indoor spaces subdivision (Kruminait e˙ and Zlatanova, 2014).

Criteria

Influence

Examples

Attractiveness of spatial element

People distribution within an area

Shop windows or artworks in galleries VS workplaces or printing corners

Necessity of spatial element

Functional areas

Information desks VS shops and cafeterias in stations

Object’s closeness to central point of the environment


Restaurants, cafeterias or shops in airports which are far away from gates, because people want the least efforts to continue their trip

Limited capacity

Contain a limited number of people

The benches in waiting halls of airports also can be seated by a limited number of people

Transition zone


To see a painting or an information screen, people stand at a certain distance from these objects

Private space

The size of the functional area of an object

Co-workers in office VS travellers in in railway stations

unit. Therefore, functional areas of objects appear in directions where services are provided, e.g., the space, in front of the information desk can be occupied by questioner, is the functional space of the desk. Moreover, certain objects might attract a larger number of people compared to others, e.g.,queue of people may appear near the before mentioned information desk while coffee table is usually seated by a certain number of people. Another example is that, In winter, people on the seats near the heater more than that without heater around. To determine the functional space of the objects in indoor space subdivision, ¯ (Kruminait e˙ and Zlatanova, 2014) put forward some criteria, examples can be seen in Table 2.2. The criteria include six aspects: attractiveness, necessity, closeness to the central location, limited capacity, transition zone, and private space. The attractiveness, closeness to the central location, and transition zone have influence on the distribution of people within an area. For instance, the restaurants, cafeterias or shops in airports which are far away from gates are less attractive to travellers, since they want to spend the least efforts to reach these places before continuing their trips. In summary, there are some shortcomings in current research on the contexts in space subdivision for pedestrian navigation. • Most of current studies in space subdivision assume that the spaces are empty. Only a few obstacles based on the specific applications are included. • Objects in indoor spaces for pedestrians has been more investigated than outdoor.

22

Chapter 2. Literature Study • The criteria for functional space determining only based on the places for specific applications (e.g., train station, airport, etc.), rather than based on the pedestrian contexts. However, the contexts decide which object should be take into account, and the number of a certain type pedestrian can have a big influence on the criteria.

23

Chapter 3

Proposed PhD Research 3.1

Problems in Pedestrian Navigation

Although a lot of research have been done in past decades, the seamless indoor and outdoor pedestrian navigation is still not very successful yet. For instance, different spatial models are dedicated for different applications, leading to these models cannot be integrated into a unified one. The current outdoor pedestrian navigation network is limited, because there is no navigation network that really belongs to outdoor pedestrian navigation, except the network adapted/reused from vehicle navigation. Without considering the space subdivision related contexts when constructing the pedestrian navigation models, pedestrians cannot find user-adaptive paths. 1. Navigation spatial models dedicated for different applications cannot be integrated directly into a unified one. A variety of approaches can be used for indoor pedestrian navigation models construction. These models constructed by using the same approach could be integrated. But if different approaches are used, models are likely to be incompatible. However, in general, the navigation models from different applications based on different approaches are different, and different applications in different spaces have different navigation models. 2. There is no navigation network that really belongs to outdoor pedestrian navigation, except the network adapted/reused from vehicle navigation. In outdoor space, there is no navigation network that really belongs to pedestrian navigation, except the network adapted/reused from vehicle navigation. This adaptation approach led to navigable space in outdoor for pedestrian navigation is shrunken or extended. For instance, pedestrians are navigated to make a detour in large space without clear paths (e.g., the square) , but the reality is that pedestrians can across the large space directly. So, the navigable space for pedestrian is shrunken. Another case is the navigable space is extended. For instance, pedestrians are navigated to walk in the car roads, which is surely not allowed considering the security of pedestrians. 3. Space subdivision related contexts are neglected. In the navigation models, most of current studies in space subdivision assume that the spaces are empty. Or only a few general objects (destinations) determined based on the specific applications, rather than the purposes of the pedestrian (utilitarian walking, rambling, strolling, promenade, or for special events). Therefore, space subdivision related contexts are neglected.

24

Chapter 3. Proposed PhD Research

To solve above three problems, this research will develop a framework, which can construct a unified 3D spatial model for space-subdivision-based pedestrian navigation in indoor and outdoor, and take into account specific contexts. Specifically, we will give clear definitions and classifications of the spaces. Indoor and outdoor spaces will be dealt with the same vision (indoor space and outdoor space are the same for pedestrian), thus using the spaces in the same way (the same structure, same management methods, as well as the same space subdivision rules, etc.). Furthermore, the context related to space subdivision will be considered. Specifically, we will determine what kind of objects should be included in space subdivision based on the types of pedestrians (contexts). Except the objects, the functional spaces of objects will be formalised. Then the combination (objects + their functional spaces) will act as the non-navigable spaces in pedestrian navigation. The criteria for functional spaces determining are also indispensable, and we plan to shared them for outdoor space subdivision.

3.2

Research Questions

There are two fundamental assumptions in this research: (a) from the pedestrian navigation perspective outdoor and indoor spaces are essentially the same, only their accessibility is restricted by different objects; (b) everywhere that can be accessed by a pedestrian must be a space, which belongs to one of the four specified categories, indoor space, outdoor space, semi-indoor space, and semi-outdoor space. With the expected research outcome, the main research question for this proposal is: What 3D space subdivision can enable context-aware seamless indoor and outdoor pedestrian navigation? To answer this main research question, six sub-questions in three sub-topics are involved: 1. Spaces. • What kind of spaces should be considered in pedestrian navigation?

• How to define, classify, and manage these indoor and outdoor spaces for pedestrian navigation? 2. Spaces-subdivision. • To which extent can we define common subdivision rules between indoor and outdoor? • How to objectify outdoor spaces, and semi-bounded for the subdivision? 3. Context related to space subdivision. • What kind of contexts are related to the space subdivision for pedestrian navigation in different situations? • What criteria should be set up to determine functional spaces for space subdivision?

3.3. Methodology

3.3 3.3.1

25

Methodology Spaces

Q1: What kind of spaces should be considered in pedestrian navigation? Method: Literature review TABLE 3.1: Literature review of pedestrian preferences for the spaces.

Situation

Like

Dislike

Kim, Kim, and Leigh (2011)

Many Korean apartment units use balconies as semi-indoor spaces by installing external windows. But the balcony spaces of some units have been removed and have been replaced in living room.

Reduce cooling and heating energy requirements, dry laundry and grow plants, by using the semi-indoor spaces.

Heat through dows.

Du, Bokel, and Dobbelsteen (2014)

Wind velocity in the semi-outdoor and outdoor spaces can improve the thermal comfort significantly.

Comfortable temperature in in the semioutdoor spaces in summer period.

Higher temperature than the comfortable temperature in summer period.

Pagliarini and Rainieri (2011b)

Compared with the outdoor sensation under the same climate conditions, the human thermal sensation inside a semi-outdoor space enclosed by a semitransparent pitched roof is better.

A better thermal sensation in semi-outdoor, which usually lower effect of wind, less hot than the outdoor space.

Bad thermal sensation in outdoor space.

Nasir, Salim, and Yaman (2014)

The semi-open space used by people for meeting, waiting, doing presentation and exhibition, doing curriculum.

Shade, and the medium ventilation.

Hot sun, and less or too much ventilation.

Lin (2009)

The number of people visiting the square increased as the thermal index value increased.

A cool temperature and weak sunlight, and adapted to thermal environments by seeking shelter outdoors.

The number of people frequenting the square decreased as the thermal index increased during the hot season.

He and Hoyano (2010)

Investigate the microclimate in a membrane structure building with a semioutdoor space during a summer period by field measurements.

Low air temperature and shelters for the sun and wind in semi-outdoor spaces.

High air temperature, hot sun and strong wind.

lose win-

26


Void Space

(a)

(b)

F IGURE 3.1: Walls in volumetric-based space model.

We will investigate this sub-question by summarising the results from literature, e.g., the research on thermal aspects. Then find out what kind of spaces should be considered in pedestrian navigation. • Studies have demonstrated that the semi-indoor/outdoor spaces have a considerable impact on pedestrians’ use of space. For example, the research on thermal aspects, from which we will try to conclude what spaces pedestrian like and dislike to visit in what situations, see Table 3.1. • From space and the size of the pedestrian perspective. I will take the book Neufert Architects’ Data (Neufert, Neufert, and Kister, 2012) as the main reference to summarize the size of the space itself, at the same time, the body measurements and space requirements of the pedestrian can give some parameters for the space choosing. Thus, we can determine what size of spaces need to be considered when building the 3D models. So as to make sure what kind of the spaces should be considered in pedestrian navigation. • Some research results about where pedestrian is willing to visit. For instance, pedestrians (travellers) like to visit somewhere with a good eyesight for taking pictures, somewhere with benches for short break, somewhere suitable for entertainment (e.g. barbecue), etc. Q2: How to define, classify, and manage these indoor and outdoor spaces for pedestrian navigation? Methods: Classify and define spaces based on the structures of the space, and then design a volumetric-based space model for management, including the schema, and data structures (tables in DBMS). We will define and classify the space based on space structures (e.g., Top structure, & Side structure, see 3.4.2 Space Classifications and Definitions), then a unified volumetric-based space model (every objects will be regarded as spaces) will be put forward to represent the spaces, see Figure 3.1, in which the walls are separate objects or cells with certain thickness (Lee et al., 2014), and pedestrians cannot have activities in them. CityGML consider walls as thin (or ‘papers’), or the walls are part of the space, see Figure 3.2. IFC basically include the name, thickness, and material information of wall components (Fazio et al., 2007). But, the IFC only for the buildings, which is not sufficient for both the indoor and outdoor pedestrian navigation. In this research, all of the physical space and objects in indoor, outdoor, semibounded space are represented as volumetric spaces, then we can get void space and occupied space. All the occupied spaces and part of the void spaces constitute

3.3. Methodology

27

F IGURE 3.2: Wallsurface/‘paper’ wall in the CityGML (Gröger et al., 2008).

F IGURE 3.3: Example of the distance between people and wall in a straight corridor (Bosina et al., 2016).

the non-navigable space, while navigable space can be extracted from the void space according to the characteristics and needs of pedestrians. Part of the non-navigable space can be regarded as void spaces base on pedestrians, e.g., pedestrians like to keep a certain minimum distance from walls and obstacles (Bosina et al., 2016), see Figure 3.3, which is a straight corridor where areas around walls, a vending point and a pillar are not used by pedestrians, illustrating the principle of separation distance.

3.3.2

Spaces-subdivision

Q1: To which extent can we define common subdivision rules between indoor and outdoor? Method: Devise some rules (algorithms) based on variable size cells to subdivide both indoor and outdoor spaces. We consider that sharing the same subdivision rules in indoor and outdoor space is feasible, because the indoor and outdoor is the same for pedestrian navigation, although the indoor environment is different from the outdoor environment. Figure 3.4 shows the possible subdivision approach for indoor space while Figure 3.5 is an example of 3D subdivision considering the terrain elevation in outdoor. We can imagine the furniture in indoor as the buildings in the outdoor spaces. In turn, we can also imagine buildings in outdoor as furniture in indoor. Therefore, the indoor space, and outdoor space can be regarded and managed in the same way. We will take all of the objects and void spaces as volumes (spaces), which have length, width, as well as height. Then, space-subdivision rules (algorithms) based on variable size cells (Nagel et al., 2010) for both indoor and outdoor will be investigated. Since variable size cells based method is more promising in computational efficiency for outdoor compared to the octree, tetrahedron, 3D voronoi diagrams, and voxels.

28


F IGURE 3.4: Possible subdivision approach for indoor space (Diakité and Zlatanova, 2017).

(a)

(b)

(c)

F IGURE 3.5: Outdoor navigation network based on space subdivision (Rusné, Ljiljana, and Beirao, 2017).

Moreover, long or irregular indoor spaces (such as corridors, concave shapes) will be further subdivided, because one node does not represent well the structure or the way of movement in the space (Brown et al., 2013). Furthermore, in this research, pedestrians are 3D objects, who have a 3D buffer space. In the vertical direction, pedestrians have height, people will make a detour where lower than their height in normal condition. Q2: How to objectify outdoor spaces, and semi-bounded for the subdivision? Method: Define some rules (parameters) based on navigation objects and space characteristics to enclose spaces for the subdivision. Our derivation of navigation network based on 3D space subdivision and Poincaré Duality requires enclosed space objects as input, but the outdoor spaces, and semibounded spaces are not as sharp as the indoor spaces. Therefore, to subdivide outdoor, and semi-bounded spaces, the first step is to find their boundaries. This research will put forward two approaches to find the space boundaries, one is based on pedestrian, and another is based on space structures. The first principle is the actual size of the navigation objects, and the functional area. Pedestrians are 3D objects, and their size is decided by the objects they carrying or using, e.g., the actual width of the pedestrian becomes the width of the cart for cleaning work. Some example of closing space for space subdivision can be seen in Figure 3.6. The size of a pedestrian(s) is: P = {Lp , Wp , Hp }

(3.1)

3.3. Methodology

29

where the Lp , Wp , Hp are the length, width, and height separately. And the size of objects (tools) for walk can be: T = {Lt , Wt , Ht }

(3.2)

E = {L, W, H}

(3.3)

where the Lt , Wt , Ht are the length, width, and height separately. Then the actual size of the navigation object is:

in which L = max{Lp , Lt }, W = max{Wp , Wt }, H = max{Hp , Ht }. I-space

O-space

sI-space

(a)

(b)

(c)

(d)

(e)

(h)

(f)

(i)

Buildings

sO-space

(g)

(j)

F IGURE 3.6: Closing spaces for space subdivision.

Another principle is based on the space itself. We will classify the space into four types: Indoor space, Outdoor space, Semi-indoor space, and Semi-outdoor space. The definitions of these spaces can be seen in the subsection 3.4.2 Space classifications and definitions. The indoor space is the space that is enclosed completely by top (solid roof) and side (solid walls) at the same time. So, we can find the boundaries of the space by detecting if the space is enclosed completely by roof and walls. The semi-indoor space has top but not physically enclosed completely by man-made or natural materials/structures/objects (e.g., wall, stone, etc.). Therefore, we can enclose semi-indoor space by its top and the objects who are supporting the top. What’s more, the patterns of space visiting also can be a very good start point to identify and find the boundaries of the Indoor space and Semi-outdoor space, see Figure 3.7. For example, passing through the door to visit the space is a must for indoor space. For the Semi-outdoor space, pedestrians have to visit an entrance or a door. However, currently, these examples are completed manually. Therefore,

30

Chapter 3. Proposed PhD Research Door

Door

I-space

Door Door

Door

sI-space

Door

Door

sO-space

Entrance/Door

Entrance/Door O-space

Entrance/Door

F IGURE 3.7: Patterns of space visiting.

automatically closing spaces for space subdivision will be investigated in this subquestion.

3.3.3

Context Related to Space Subdivision.

Q1: What kind of contexts are related to the space subdivision for pedestrian navigation in different situations? Method: Investigate space subdivision related contexts, then define criteria (e.g., attractiveness) based on theses contexts. We will investigate space subdivision related contexts by literature review. According to research Great Streets Initiative1 , there are five reasons why people travelling as pedestrians in both indoor and outdoor spaces, so we will classify pedestrian into five kinds: Utilitarian Walking pedestrian, Rambling pedestrian, Strolling/ Lingering pedestrian, Promenade pedestrian, and Special Events pedestrian. Then, the objects that may appear in the navigation of each type of pedestrian will be summarized. For instance, in a university, staffs may walk as utilitarian walking pedestrians. The objects in their navigation (based on their contexts) could be printers, coffee machines, toilets, projectors, coffee tables, etc. Q2: What criteria should be set up to determine functional spaces for space subdivision? Method: Formalise the criteria (e.g., attractiveness during the time by peak hour, and off-peak hour) for space subdivision. We will improve the value range of the current criteria, and estimate the criteria in quantity ways by adapting some mathematical models. Currently, we will focus ¯ ˙ 2014), including on six criteria for the space subdivision presented in (Kruminait e, attractiveness, necessity, closeness to the central locations, limited capacity, transition zone, and private space, see Table 3.2. The value range is not very accurate, e.g., the attractiveness. With only three levels, non-attractive, moderately attractive, and highly attractive, it will be very ambiguous when the attractiveness is about 80% or 35%. We cannot tell this case should belong to which level. Another example is the limited capacity, which value ranges are only yes or no. Actually, any object has a limited capacity, or a unlimited capacity. For instance, a long bench without clear number of seats, which can hold many people can even other people can sit on 1 http://www.greatstreetsstlouis.net/residential-neighborhood-design/multimodal-corridorplanning/pedestrians

3.3. Methodology

31

the ground around the bench, or it can be occupied only one people who is sleeping. Therefore, the criteria decided by many factors, rather than only the physical features. An example of estimating the criteria in quantity ways by adapting some math¯ ˙ 2014). it is adopted from (Sevtematical models is that the closeness in (Kruminait e, suk and Mekonnen, 2012) where using network analysis method closeness of an input features is defined as the inverse of cumulative distance required to reach from specific node to all other nodes that fall within the search radius along the shortest paths. In other words it can be said that closeness of an object describes how far the object is from its surrounding neighbours. Therefore, closeness is estimated as: Closenessr [i] =

1 ⌃j2G

{i},d[i,j]r (d[i, j])

(3.4)

where the Closenessr [i] is the closeness of node within the search radius r, d[i, j] is the shortest path distance between nodes i and j. TABLE 3.2: Criteria and value ranges of functional spaces in indoor ¯ ˙ 2014). space (Kruminait e,

Criterion

Measurement

Value range

Attractiveness

How inviting is the structure of the object?

1 - non-attractive; 2 - moderately attractive; 3 - highly attractive

Necessity

Is it necessary to have this object in this environment? Is it an important/essential feature of the environment?

0 - non-essential object; 1 - essential object

Object’s closeness to central locations

How close object is to all other surrounding objects?

[0-1] 0 - object is far away from other locations; 1 - object is close to other locations

Limited capacity

Does the object have limited number of seats?

Yes - object has limited capacity; No - object does not have limited capacity

Transition zone

Does the object provide services in a distance?

Numerical variable based on structure of the environment

Private space

What is the minimum distance that people keep in order not to violate others personal space in this environment?

Numerical variable based on type of the building (Hall’s personal, social distances)

Another example is the attractiveness of the information desk, see Figure 3.8. The functional (coloured) areas around the information desk are not non-attractive, moderately attractive, and highly attractive only, but can be calculated by some mathematical models. We plan to model it into four areas (every 0.25 meter, blue, green, yellow, and red) within 1 meter, since 1 meter is taken as a rule in a lot of places, where they set a mark to remind people to wait outside this line. But it can be less if there are a lot people want the information desk with some special events.

32


F IGURE 3.8: Information desk with functional areas (coloured areas).

IFC/BIM

CityGML/IndoorGML

Volumetric-based space model

Functional Spaces

Database Management System

Criteria Object Space

Free Space

Navigation Network

Unoccupied Space

Navigation Nodes

∩ Occupied Space

Navigation object (Pedestrian)

Poincare Duality

Contexts

Inaccessible Space

Accessible Space

Sub-spaces

Non-navigable Space

Navigable Space

Algorithms

F IGURE 3.9: Workflow of my research.

Moreover, more criteria will be investigated except for the above six, especially taking the pedestrian contexts into consideration, e.g., the body measurements and space requirements (Neufert, Neufert, and Kister, 2012), the number of the pedestrians - whether only one pedestrian is needed to be navigated.

3.3.4

Implementations & Case Study Models

Figure 3.9 shows the workflow of this research. We plan to use IFC/BIM, CityGML/IndoorGML as the data sources, and tailors them to fit the space computations, because these are the related prevailing international standards. What’s more, the methods of constructing 3D indoor and outdoor geometry models are not included in this research, but I will construct some 3D geometry models manually to prove the concepts that I will develop. More importantly, we will make full use of semantics in these models, rather than assign semantics to objects manually, because it will be very complex and time consuming. To be specific, all of the data (3D geometry space models, as well as navigation models) will be managed in the database management system by designing spatial tables. The input data (IFC/BIM, CityGML/IndoorGML) will be tailored to meet the needs of space management, for instance, using the method presented in (Diakité and Zlatanova, 2016b) to generate valid indoor spaces in IFC models. We can get the object space directly, then the free space can be extracted semantically rich

3.3. Methodology

33

I-space

sI-space

Water

Vehicle lanes

O-space

sO-pace

Vegetation

Buildings

B6

P3

B7 G1

B8

B4

B5

B9

B10

P2 B14

B13

B15

B16

B3

P1 B12

B11

B17 (a)

R1 B2

B1

(a)

(c)

(b)

(d)

F IGURE 3.10: (a) is the 2D map of the imaginary small city; (b) is the 3D map of the navigation area; (c) is the spaces is pedestrians can use; (d) is the sI-spaces and sO-spaces.

and furnished IFC models (Diakité and Zlatanova, 2016a). The functional space can be determined from object space based on some criteria, e.g., attractiveness. If the functional space is using (occupied), it becomes the non-navigable space. While the unoccupied space (free space) can be separated into inaccessible space and accessible space based on the navigation object (pedestrians, or pedestrian with some tools). Based on the the kind of pedestrian, navigable space will be subdivided into sub-spaces, and some sub-spaces can be occupied spaces. The navigation nodes and links can be derived from sub-spaces based on the Poincaré Duality theory. The final navigation model also will be stored and managed by the database management system. An imaginary small city, including indoor and outdoor, will be taken for case Faculty of Architecture and the Built Environment

Civil Engineering and Geosciences

F IGURE 3.11: Planned experimental scenarios in TU Delft.

34


study models, see Figure 3.10. The Figure (a) and (b) are the 2D map and 3D map of this area. More specifically, there are 17 buildings (B1 B17), 3 pavements (P 1 P 3), 1 garden with the fence (G1), a road (R1) for vehicles, and a river. Three buildings (B2, B12, and B17) contain eaves. Moreover, B12 also has a yard with fence. The building B11 is a courtyard. The spaces, I-space, O-space, sI-space, and sO-space, are coloured with white, light yellow, light green, and blue respectively, see Figure (c). The semi-spaces we extracted them separately in order to show them clearly, see the spaces in Figure (d). What’s more, I will try to test this research in the TU Delft, see Figure 3.11. We will choose the building of Faculty of Architecture and the Built Environment and the building of Civil Engineering and Geosciences as the indoor spaces, and the outside spaces between them are outdoor spaces.

3.4 3.4.1

Preliminary Results Pedestrian definition

There are many definitions about the pedestrian, in the Wikipedia2 , a pedestrian is a person travelling on foot, whether walking or running. In some communities, those travelling using tiny wheels such as roller skates, skateboards, and scooters, as well as wheelchair users are also included as pedestrians. In modern times, the term usually refers to someone walking on a road or pavement. In the Oxford Dictionaries3 , the pedestrian is a person who is walking rather than travelling in a vehicle. In the Merriam-Webster4 , pedestrian is going or performed on foot.

Pedestrian Walking

Running

Robot walking or running on foot

Pedestrian + Tiny wheel tools

People with scooter

Roller skating people

Skateboarding people

Pedestrian + Wheel tools Walking with trolley

Self-balancing scooter

Self-balancing vehicle

F IGURE 3.12: Proposed definition and types of the pedestrians.

In this research, we define the pedestrian as a person travelling on foot, whether walking or running. Or a robot who can mimic people, also travelling on foot. As for those persons or robots, they are not pedestrians when travelling using tiny wheels such as roller skates, skateboards, and scooters, or wheels such as self-balancing scooter, because they have lost the flexibility of pedestrian movement, for instance, pedestrians can walk on stairs, but those persons or robots have to quit from these tiny wheel tools or wheel tools and back to pedestrian (travelling on foot). Therefore, the pedestrian without any tools have the maximum flexibility from the point of view of the pedestrian navigation, but with the tiny wheel tools or wheel tools, they 2

https://en.wikipedia.org/wiki/Pedestrian https://en.oxforddictionaries.com/definition/pedestrian 4 https://www.merriam-webster.com/dictionary/pedestrian 3

C T (%)

3.4. Preliminary Results

35

0 ≤ α ≤ η ≤ β ≤ 100 & 0 ≤ γ ≤ δ ≤ 100

100

sI-space

I-space

O-space

sO-space

β

η α

0

γ

δ

100

C S (%)

F IGURE 3.13: Features of the spaces.

prefer to move on flat space, while for the tools with wheel, the flatter the better, see Figure 3.12.

3.4.2

Spaces Classifications and Definitions

The definition of space shares four unanimously approved characteristics: boundless, extensible, three-dimensional, and can be occupied. But this general space definition is not sufficient from the perspective of pedestrian navigation. In the above review, it is determined that space that is classified as indoor or outdoor is more suitable for pedestrian navigation. Moreover, the core of the space for pedestrian navigation is the unoccupied parts, where people can have activities, behaviours, etc., rather than shells (man-made or natural materials/structures/objects), who are surrounding the unoccupied parts. In this study, four kinds of spaces are included in our framework, i.e., space is formally classified as indoor space (I-space), outdoor space (O-space), Semi-indoor space (sI-space), and Semi-outdoor space (sO-space). S=I

[

O

[

sI

[

sO

(3.5)

where S, I, O, sI, and sO denote Space, I-space, O-space, sI-space, and sO-space respectively. Spaces are classified based on seven parameters: two physical structure parameters, Top closure (C T ), and Side closure (C S ), and five closure parameters (↵, , , , and ⌘), see Figure ??. Physical structure parameters are invariable, while all of the closure parameters may vary based on the specific situations in pedestrian navigation. This means, the parameters are determined in two steps: (a) Physical Structure computation. C T , and C S are calculated based on the presence of physical structures on the top and sides of a space by the Equation 3.6. (b) Closure parameters estimation. ↵, , , , and ⌘ are estimated based on the user’s situation. For instance, if a pedestrian needs to escape rain, he or she would prefer to pass as many as I-spaces and sI-spaces as possible, which are likely to have shelters from the rain. In this case, all spaces that could be visited can be found based on the route’s start and end locations. Then the Physical Structure parameters of these spaces can be calculated respectively. Closure parameters are decided based on the pedestrian’s specific situation and preferences.

36


(a) Roof

(b) Top

(c) Wall

(d) Side

F IGURE 3.14: Example of the roof, top, wall and side.

Physical structure parameters depend only on the spaces themselves, while closure parameters are based on specific situations in pedestrian navigation. Specifically, these parameters can be determined by two steps: (a) Physical closure computation. C T , and C S can be calculated based on the physical structures of the top and side by the Equation 3.6. (b) Closure parameters estimation. ↵, , , , and ⌘ are estimated by the user’s situations. For instance, if a pedestrian wants to escape from the rain, he can choose to pass as many as spaces with good shelters (I-spaces and sI-spaces) to escape from the rain. Therefore, for this case, the spaces could be visited can be determined by his start and end location, then the physical structure parameters of these spaces can be calculated respectively. Closure parameters are estimated by his specific situation. For instance, he can set them as: ↵ = 5, = 95, = 5, = 95, and ⌘ = 80. Since in rainy day, he thinks that only the I-spaces (C T > 95% & C S > 95%), or the sI-spaces (C T > 80%), are ideal spaces for him to escape the rain. But he may consider that the ↵, , , and can be set as 20, 60, 20, 65, and 50 respectively are enough to avoid the hot sun. The definitions of Roof, Wall, Top, Side, Top closure, and Side closure are following. Roof5 : A roof is part of a building envelope. It is the covering on the uppermost part of a building or shelter which provides protection from weather, notably rain or snow, but also heat, wind, and sunlight. The word also denotes the framing or structure which supports that covering. Wall6 : A wall is a structure that defines an area, carries a load, or provides shelter or security. The purposes of the walls in buildings are to support roofs, floors, and ceilings; to enclose a space as part of the building envelope along with a roof to give buildings form; and to provide shelter and security. 5 6

https://en.wikipedia.org/wiki/Roof https://en.wikipedia.org/wiki/Wall


37 I-space Yes

Yes

Spaces

I-space or sI-space

High C T & High C S

No

sI-space

CT No

O-space or sO-space

T Low CYes & Low C S

No

sO-space

Yes O-space

F IGURE 3.15: The workflow of proposed spaces definition framework.

Top: Top is above the ground and has the capability for people to have activities under it. Roof is always a top, but Top is not always a roof. This is because Top does not always provide protection from weather. For instance, some tops can help people escape sunlight, but not rain. See the red rectangle parts in Figure 3.14, in which the (a) is Roof, while (b) is Top. Moreover, differently than Rood which is always a man-made structure, Top can also be natural (e.g., stone, tree canopy). Side: Side is a structure that encloses an area. To a certain extent, a side is more like a fence, although it may have the similar functions of a wall, such as, carrying a load (e.g., rows of load-bearing pillars). In general, a side is used to draw boundaries, to prevent people enter or exit. See the blue rectangles in Figure 3.14, in which the (c) is Wall, while (d) is Side. A wall is one kind of side, but a side maybe not be a wall. Moreover, a side can be man-made structures (e.g., wall, fence, etc.) or natural structures (e.g., tree, river). Top closure (C T ): top closure means the proportion of the top closing, i.e., the area of the top enclosed part is divided by the total area of the top, see Equation 3.6. Side closure (C S ): side closure means the proportion of the side closing, i.e., the area of the side enclosed part is divided by the total area of the side, see Equation 3.6. c = A_c/A_t

(3.6)

in which the c is C T or C S , while A_c and A_t are the area of enclosed part, and the total area of the Top/Side respectively. There are two fundamental assumptions in this research: 1) from the pedestrian navigation perspective outdoor and indoor spaces are essentially the same, only their accessibility is restricted by different objects; 2) every place that can be accessed by a pedestrian must belong to one of the four above specified categories. The workflow of the proposed definition framework based on C T and C S can be seen in Figure 3.15. For the spaces, C T is the first criteria. If a space has a Top, it can be I-space or sI-space. Then the C T and C S together is the second criteria. If space has high physical C T and C S at the same time, it is I-space, otherwise, it is sI-space. Space without the C T , can be O-space or sO-space. If space has low physical C T and C S meanwhile, it is O-space, otherwise, it is sO-space. It is important to mention that category of a space entirely depends on the top (closure) and side (closure). The with/without of C T in the first criteria means a certain level is based on the situations, i.e., for a certain case, the closure ↵% is enough. Then if the C T ↵%, it means the space has top, otherwise it is without a top. High/low physical closure of Top/Side is similar, but it depends on the situation.

38


Environment Definition

Indoor

CT ≥ β% & C S ≥ δ %

Semi-indoor

Semi-outdoor

C T ≥ η% & C S ∈[0,1]

C T < η% & C S ∈[0,1]

except C ≥ β % & C ≥ δ % except C T < α % & C S < γ % T

S

Outdoor

CT < α % & C S < γ %

Example

Scene

Parameters

0 ≤ α ≤ η ≤ β ≤ 100 & 0 ≤ γ ≤ δ ≤ 100

F IGURE 3.16: Definitions of the spaces based on proposed definition framework.

Figure 3.16 shows how spaces can be classified using the above described framework. • Indoor space is a space with both Top (solid roof) and Side (solid walls) at the same time, in which the C T is more than %. Meanwhile the C S is more than %, see the red portion in Figure 3.13. • Semi-indoor space is a space with a top; but sides are not essential. In particular, the C T cannot be less than ⌘%, while the C S can be any value from 0 to 100%. More importantly, the indoor situation is not included; see the orange portion in Figure 3.13. • Outdoor space is the space where the C T is less than ↵%, meanwhile the C S less is than %, see the blue portion in Figure 3.13. • Semi-outdoor space is a space with C T that must be less than ⌘%, while C S is still more than %. Meanwhile the space with C S is between ↵% and ⌘%, but, it is also semi-outdoor space if the C T is between 0 and %, see the green portion in Figure 3.13. Characteristics and relationships of spaces based on closures can be seen Figure 3.13. Even the same structure with the same function can belong to different categories defined in above, see bridges between two buildings in Figure 3.17. (a) is I-space, but (b) is sI-space, while (c) is sO-space, because the bridge spaces in (a) and (b) have tops and sides. But the top closure and side closure of the former are nearly 100%, while the side closure of the latter is at most 50%. Therefore, the former is indoor, and the latter is semi-indoor. The bridge space in (c) has no top, and it has a side closure more than 90%, so it is sO-space. Figure 3.18 shows more examples of space classification, in which the names are as follows: (a) Room; (b) Open air; (c) Roof terrace with top; (d) Gas station; (e) Under the overhanging roof part; (f ) Under the overhanging roof part; (g) Under an overpass; (h) Porch; (i) Bus stand; (j) Courtyard; (k) Roof terrace with sides; (l) Roof terrace with outside stairs; (m) Yard with fence; (n) Road with side fence; (o) Stadium.


(a) Indoor

39

(b) Semi-indoor

(c) Semi-outdoor

F IGURE 3.17: Bridges between two buildings.

In this case, we set ↵, , , , and ⌘ as 5, 90, 5, 90, and 80 respectively. Base on proposed definition framework, the Figure 3.18(a) is an I-space, while the space in Figure 3.18(b) is O-space, because both the top closure and side closure in (a) are more than 95%, but the second scene with top closure and side closure less than 5% is open air. Figure 3.18(c), 3.18(k), and 3.18(l) are roof terraces. The first one is sI-space because its top closure over 90%, in contrast the second and third are semi-outdoors since they do not have a top, but they have almost 100% side closure. Because of the roof with 100% top closure, the gas station Figure 3.18(d), under the overhanging roof part Figure 3.18(e) and Figure 3.18(f), under the overpass Figure 3.18(g), porch Figure 3.18(h), and bus stand Figure 3.18(i) are sI-spaces. However, the courtyard Figure 3.18(j), although surrounded by adjacent buildings, it is a sO-space because it the top is not closed, but it has a side closure more than 5%. Moreover, the yard with fence in Figure 3.18(m), and the road with side fence in Figure 3.18(n) are sO-spaces. The football court within the stadium in Figure 3.18(o) is sO-space, since it has no top, but it has a side with the C S more than 5%. However, the spaces under the bridges are sI-spaces, because they have a top with the C T more than 90%.

3.4.3

Space Management

In this research, all of things are spaces (Volumes), see Figure 3.19. The general notion of the space is Spaces, which includes Indoor Space, Outdoor Space, Semiindoor Space, and Semi-outdoor Space. Void Space can exist in all the four types of spaces. In Indoor Space, Floor, Furniture, and Wall are Non-Navigable Space, while Door, Window, Connection, and Void Space are Navigable Space. The Floor is a separate space, which cannot be through, but Furniture can stand on. The Connection is a general concept of objects who can connect different spaces, such as Stair, Escalator, Elevator, and Lift. In Outdoor Space, Ground, CityFurniture, Waterbody, and some of Vegetation (e.g., trees) are Non-Navigable Space, but some of vegetation (e.g., lawn) are Navigable-space, in which the Ground is also a separate space, and other space can be on or in it, e.g., Building can inside the Ground or on the Ground. Bridge, Tunnel, as well as Building are Navigable Space, although sometimes, they can be closed because of maintains, or emergency situations. Moreover, Navigable Space are Changeable Space, since it can be changed into Non-Navigable Space, for example the Door is closed. Window also is Non-navigable space in non-emergency situations. Semi-indoor/outdoor Space is a bit of special, and only Void Space is considered.

40


(a) Indoor

(b) Outdoor

(c) Semi-indoor

(d) Semi-indoor

(e) Semi-indoor

(f) Semi-indoor

(g) Semi-indoor

(h) Semi-indoor

(i) Semi-indoor

(j) Semi-outdoor

(k) Semi-outdoor

(l) Semi-outdoor

(m) Semi-outdoor

(n) Semi-outdoor

(o) Semi-outdoor

F IGURE 3.18: Some examples of the spaces.


41 Spaces + absoluteHeight:float

Indoor Space

Ceiling (volumetric) Floor (volumetric)

Semi-indoor Space

Wall (volumetric)

Furniture (volumetric)

Semi-outdoor Space

Bridge (volumetric)

Connection (volumetric) Window (volumetric)

Stair (volumetric)

Outdoor Space

Tunnel (volumetric)

Building (volumetric)

CityFurniture (volumetric)

Vegetation (volumetric)

WaterBody (volumetric)

Grass (volumetric)

Escalator (volumetric)

Tree (volumetric)

Elevator (volumetric) Lift (volumetric)

Door (volumetric)

Convertible Space (volumetric)

Void Space (volumetric)

Non-navigable Space (volumetric)

Navigable Space (volumetric)

Volumetric + Horizontal + Vertical + Gradient 1..* Coordinates +x: float +y: float +z: float

Vertex 1

0..1

+ elevation: float

Edge 2

1

+ start: Vertex + end: Vertex

ConvexSpace

Triangle 3

0..2

+edges: Edge

1..*

1

+ triangles: Triangle [1…] + edges: Edge [3…]

F IGURE 3.19: The volumetric-based data model of spaces.

3.4.4

Experiments of Enclosing Spaces

We did experiments on how to find boundaries of space for 3D subdivision, in which the inputs are the 3D models of the imaginary small city, and all of data processes are completed automatically by Rhino + Grasshopper (Python script). Roof is a critical element for the indoor space, and semi-indoor space. Because of different shapes of the roof, the spaces can be different shapes as well. Therefore, the experiments start from the space with roof. After investigating the roof shapes (Figure 3.20), we demonstrated a few buildings based on their common features. Based on the roof shapes, we designed several cases to demonstrate them. The buildings components in CityGML (Gröger et al., 2008), BIM model 7 can have different names, but four parts are the basic element for a building. Therefore, I designed the inputs of my algorithm as Roof, Floor, FloorSlab, and Wall, which means that a building consists four components, i.e., Roof, Floor, FloorSlab, and Wall. More importantly, I designed two rules in my algorithm: find the surface with maximum area from each Brep object. If there is only one surface with maximum area, I will project it to the floor or ground. But if there are two surfaces have same area, and both of them are maximum surface, I will choose the lower surface to make projection. But it failed to meet the above example, so the rules improved as always trying to find top two surfaces with maximum areas. If their areas are almost same (within 7

http://www.buildingsmart-tech.org/ifc/IFC4/Add2/html/

Ground (volumetric)

42


a tolerance), I will choose the lower surface to project to the floor/ground. If they are not almost same, both of these two surfaces will be projected to the floor/ground, thus three spaces can be created. Moreover, the vertical roofs should be taken into account. Verge Ridge

min 2% fall

Junction point Ridge Hip

Gable

(1) Flat roof

(2) Single-pitch (monopitch) roof

(5) Half-hipped roof

(6) Mansard roof

(9) Two single pitches

(10) Northlight or saw-tooth roof

(11) Four hables

(14) Gabled dormer window

(15) Roof cut-out

(13) Wide dormer with sloping roof

(4) Hipped roof

(3) Gabled roof

(8) Coumpound roof with central gutter

(7) Barrel roof

(12) Square hipped roof

(16) Single pitched-roof dormers

A

B

C

A

B

C

Spaces

Original 3D model

F IGURE 3.20: The basic forms of roofs and roof projections (Neufert, Neufert, and Kister, 2012).

F IGURE 3.21: Three examples of space creating.

Figure 3.21 illustrated tree different buildings with different roof shapes. Building A is a four-storey apartment with flat roof. Building B is a house with two gabled roofs, and one flat roof. Building C is a courtyard building with flat roof. Based on


43

previous definition and rules, four storey spaces can be created in Building A, further these four spaces can be cut into eight spaces, four indoor spaces, and four semi-indoor spaces. Obviously, building B has four (or one) indoor space(s), and one semi-indoor space (the space under the flat roof). Spaces in building C are a combination of indoor space, semi-indoor space, and semi-outdoor space.

(a) original 3D storey model

(b) Walls

(c) Floor

(d) Outer walls

(e) Inner walls

(f) Whole space of the storey

(g) Inner walls as cutters

(h) Room spaces

(i) Top view of room spaces

F IGURE 3.22: Inner walls cut the floor space into room spaces.

After creating the spaces based on the boundaries, the spaces will be subdivided based on the inner walls, see Figure 3.22, in which (a) is the original 3D storey model. (b) is the walls. (c) is the floor space. (d) is all of outer walls, while (e) is all of the inner walls in the scenario, including door spaces. Based on the floor and the height of walls, the whole space of the storey can be created, see (f ). Then the inner walls can act as cutters (g) to cut the storey into room spaces, see (h) and (i). However, current experiments are some preliminary attempts and they have many disadvantages. For instance, the semantics of the objects have to be assigned manually, because the original data is 2D floor plans, in which only geometry are included. Another shortcoming is that the current data is very small, so it can be constructed manually, but for large scene (e.g., all objects in campus of the TU Delft), it will be very time consuming to get the 3D volumetric-based model. I plan to construct only two building with indoor structure, (the building of Faculty of Architecture and the Built Environment, and the building of Civil Engineering and Geosciences) and demonstrate the navigation paths.

45

Chapter 4

Practical Issues 4.1

Courses & Supervision TABLE 4.1: The courses in the first year.

Discipline-related skills Courses

Start Date

End Date

Credits

Status

14/11/2016

30/01/2017

5 EC

Completed

16/11/2016

01/02/2017

5 EC

Completed

Courses

Start Date

End Date

Credits

Status

Problem-Solving & Decision-Making in Research Presenting Scientific Research Developing Your Academic Skills: Critical/Analytical Thinking and Scientific Reflection

07/12/2016

14/12/2016

1.5

Completed

23/01/2017 03/05/2017

02/02/2017 17/05/2017

3.0 1.5

Completed Completed

Start Date

End Date

Credits

Status

12/01/2017 24/01/2017 18/11/2016 09/01/2017

12/01/2017 24/01/2017 18/11/2016 16/01/2017

0.5 1.0 0.5 2.0

Completed Completed Completed Completed

14/03/2017 19/05/2017 06/06/2017

18/04/2017 15/06/2017 16/06/2017

3.0 1.5 2.0

Completed Completed Completed

GEO1006 Geo Database Management Systems GEO1003 Positioning and Location Awareness

Research skills

Transferable skills Courses PhD Start-up Module A PhD Start-up Module B PhD Start-up Module C Self-Awareness and autonomy in the research process The Art of Presenting Science Time Management How to Interact Effectively with Your Research Team

To complete the Doctoral Education programme, Jinjin Yan must earn 45 GS credits (1 GS credit = 8 hours of coursework + 4 hours of preparation/assignment). The programme is divided into three skills categories. The total 45 GS credits are divided equally among these categories: Discipline-related skills (15 GS credits) The discipline-related skills category focuses on giving him a greater breadth and depth of knowledge in the field of your doctoral research. Research skills (15 GS credits) The focus of the research skills category is to improve his ability to conduct scientific research, and improve the skills needed for a role as a researcher in an academic

46

Chapter 4. Practical Issues

environment. Transferable skills (15 GS credits) The transferable skills category focuses on personal and professional development, which will help him now and in his future career. Jinjin Yan will follow 12 courses (Table 4.1) at the first year, including 2 courses from MSc programmes, Geo Database Management Systems and Positioning and Location Awareness, and 10 courses from Graduate school. If all courses are qualified, Jinjin will get 26.5 GS credits. Jinjin will register Popular Scientific Writing (C5.M3), Writing a Dissertation (C13.M1), Writing a Scientific Article in English (C12.M4), and Creative Writing: Overcoming Writer’s Block (C12.M3) to improve the English writing skills and get 18.5 GS credits at least. Moreover, Prof. Jantien Stoter will act as the promotor and spend approximately 35 hours of supervision per year. On average, we will meet and have a communication about the research at least once per month. Dr. Abdoulaye Diakité acts as the (co)promotor and daily supervisor. She will discuss with Jinjin Yan once per week about progress and problems of current research and the total time on the supervision is approximately 70 hours per year. In average, Abdoulaye and the Jinjin will meet at least biweekly. Before each weekly or monthly meeting, Jinjin will prepare an agenda and submit a Weekly Progress Monitor report to the supervisors by email. Jinjin will take notes for both monthly and weekly meetings and summarise them as minutes and actions every time. During the sabbatical leave of the supervisors, the monthly or weekly meetings will take place by telephone or E-mail contact.

4.2

Software, Tools and Data

According to the objectives and methodology of this study, the following skills/tools are required: - Programming skills - Existing Standards etc. - Data Processing - Data Storage - GIS Skills

C++, C, Python, and Java etc. ISO, and OGC (GML, CityGML, IndoorGML), IFC/BIM

Matlab, SPSS, SQL etc.

Oracle Spatial, PostGIS etc. FME, ArcGIS etc.

- Model Generation and Visualization Grasshopper etc.

AutoCAD, 3Ds Max, Sketch Up, Rhinoceros,

- Datasets An imaginary small city; The campus of TU Delft, and the indoor spaces will include the building of Faculty of Architecture and the Built Environment and the building of Civil Engineering and Geosciences. - Others Visualization devices (e.g. mobile platforms), communication tools (wireless to mobile platforms) and indoor positioning equipment.

47

Chapter 5

Deliverables 5.1

Outcomes and Reports

It is expected that a number of articles and papers will be prepared for international journals and conferences during the whole phases: • 2 or 3 papers per year in (international) conferences. e.g. ISPRS, UDMS, Gi4DM, SDH, 3D GeoInfo etc. • 2 or 3 papers in reviewed scientific journals. Potential conferences and Journals e.g. IJGIS, Sensors, GeoInformatica, Computers, Environment and Urban Systems, etc. Except for the output in the form of articles and papers, other types of outputs are to be anticipated from this research: A framework to construct 3D spatial navigation model for pedestrian in indoor and outdoor, in which some contexts will be taken into account. Specifically, • A classification and definition method for spaces. • Volumetric-based space model, schema, and data structures (tables in DBMS). • The space identification methods taking several aspects of pedestrian, such as speed, dimension, flexibility, behaviours into consideration. • Space-subdivision rules (algorithms) based on variable size cells for both indoor and outdoor. • Rules (parameters) to enclose outdoor spaces for the subdivision. • Criteria (e.g., attractiveness) for the space subdivision. • Algorithms to formalise the criteria(e.g., the time by peak hour, and off-peak hour) for space subdivision.

5.2

Time Planning

This research will be conducted over a period of 4 years. The following schedule is just a rough one. It is a time arrangement on the PhD research, which officially started on October 15th, 2016. The actual time may be adjusted as the research progresses. Figure 5.1 provides an overview of the planning activities that have been carried out and are to be done in this research. The research themes presented in the time planning overview are not strictly sequential, and will be investigated with a high degree of overlap in time. The plan is subject to change according to the

48

Chapter 5. Deliverables

flow of the research. Several tasks and activities will be a part of this research: Initial literature review. Writing proposal, Go/No Go meeting, Spaces definition & Management, Spaces-subdivision for spaces, Context definition & Management, Landmark-based guidance, Case studies, and Formalisation & Writing dissertation. Time and Items

2016 - 2017 Q1

Q2

Q3

Q4

2017 - 2018 Q1

Q2

Q3

Q4

2018 - 2019 Q1

Q2

Q3

Q4

2019 - 2020 Q1

Q2

Q3

Q4

Initial literature review Early-stage preparations

Writing proposal Go / No Go meeting A classification and definition method for spaces

Space

The space identification methods based on literature review Volumetric-based space model, schema, and data structures

Space subdivision

Criteria

Space-subdivision rules based on variable size cells Rules (parameters) to enclose outdoor spaces for the subdivision Algorithms to formalise the contexts for space subdivision Criteria (e.g., attractiveness) for the space subdivision

Prove of concept & Implementations & Case studies dissemination Formalisation & Writing dissertation

F IGURE 5.1: Time planning overview.

5.3

Short term first year plan

This section presents the proposed key events and milestones achieved in the first year of research. The first year besides writing proposal includes writing two general research papers and attending four events. Table 5.1 enlists the key events of the first year of research. TABLE 5.1: Key events in the first year of research.

Month

Key Event

2016-10 2016-11 2017-02 2017-03 2017-04 2017-05 2017-06 2017-07 2017-09

Start of research GEO1003 & GEO1006 Preliminary results 1 Writing abstract for ISPRS Workshop 2017 (accepted) Preliminary experiments Writing full paper for ISPRS Workshop 2017 Writing draft of the journal paper Preliminary results 2 Poster for ISPRS Workshop 2017 & Go/No go meeting

5.4

Publications

In the first year, two publications will be submitted. Challenges in Flying Quadrotor Unmanned Aerial Vehicle for 3D Indoor Reconstruction, Jinjin Yan, Nives Grasso, Sisi Zlatanova, Robin Christian Braggaar and Danny Marx. ISPRS Workshop Indoor 3D 2017. A Comprehensive Space-Definition Framework for Pedestrian Navigation, Jinjin Yan, Sisi Zlatanova, Abdoulaye Diakité, Jantien Stoter. (Submitted).

49

Reference Afyouni, Imad, Ray Cyril, and Claramunt Christophe (2012). “Spatial models for context-aware indoor navigation systems: A survey”. In: Journal of Spatial Information Science 1.4, pp. 85–123. Afyouni, Imad, Cyril Ray, and Christophe Claramunt (2010). “A fine-grained contextdependent model for indoor spaces”. In: Proceedings of the 2nd acm sigspatial international workshop on indoor spatial awareness. ACM, pp. 33–38. Amutha, B and Karthick Nanmaran (2014). “Development of a ZigBee based virtual eye for visually impaired persons”. In: Indoor Positioning and Indoor Navigation (IPIN), 2014 International Conference on. IEEE, pp. 564–574. Basiri, Anahid et al. (2016). “Seamless pedestrian positioning and navigation using landmarks”. In: The Journal of Navigation 69.1, pp. 24–40. Becker, Thomas, Claus Nagel, and Thomas H Kolbe (2009a). “A multilayered spaceevent model for navigation in indoor spaces”. In: 3D geo-information sciences. Springer, pp. 61–77. — (2009b). “Supporting contexts for indoor navigation using a multilayered space model”. In: Mobile Data Management: Systems, Services and Middleware, 2009. MDM’09. Tenth International Conference on. IEEE, pp. 680–685. Berg, M. de et al. (2013). “Geometric Data Structures for Games: Geometric Graphs”. In: https://www.cs.umd.edu/class/spring2013/cmsc425/Lects/lect09.pdf Lecture 9. Boguslawski, Pawel and Christopher Gold (2009). “Construction operators for modelling 3D objects and dual navigation structures”. In: 3D Geo-Information Sciences, pp. 47–59. Borovikov, Igor (2011). “Navigation graph generation”. In: Artificial Intelligence 3, p. 20. Bosina, Ernst et al. (2016). “Avoiding Walls: What Distance Do Pedestrians Keep from Walls and Obstacles?” In: Traffic and Granular Flow’15. Springer, pp. 19–26. Bouyer, J et al. (2007). “Thermal comfort assessment in semi-outdoor environments: Application to comfort study in stadia”. In: Journal of Wind Engineering and Industrial Aerodynamics 95.9, pp. 963–976. Brown, Gavin et al. (2013). “Modelling 3D topographic space against indoor navigation requirements”. In: Progress and New Trends in 3D Geoinformation Sciences. Springer, pp. 1–22. Cao, Bin et al. (2017). “Thermal comfort in semi-outdoor spaces within an office building in Shenzhen: A case study in a hot climate region of China”. In: Indoor and Built Environment, p. 1420326X17728152. Cheng, Jiantong et al. (2014). “Seamless outdoor/indoor navigation with WIFI/GPS aided low cost Inertial Navigation System”. In: Physical Communication 13, pp. 31– 43. Chengappa, Chaya et al. (2007). “Impact of improved cookstoves on indoor air quality in the Bundelkhand region in India”. In: Energy for Sustainable Development 11.2, pp. 33–44.

50

REFERENCE

Corbetta, Alessandro et al. (2016). “Asymmetric pedestrian dynamics on a staircase landing from continuous measurements”. In: Traffic and Granular Flow’15. Springer, pp. 49–56. Daamen, Winnie, PHL Bovy, and SP Hoogendoorn (2002). “Modelling pedestrians in transfer stations”. In: Pedestrian and evacuation dynamics, pp. 59–73. Diakité, Abdoulaye A and Sisi Zlatanova (2016a). “EXTRACTION OF THE 3D FREE SPACE FROM BUILDING MODELS FOR INDOOR NAVIGATION.” In: ISPRS Annals of Photogrammetry, Remote Sensing & Spatial Information Sciences 4.2, pp. 241– 248. — (2017). “Spatial subdivision of complex indoor environments for 3D indoor navigation”. In: International Journal of Geographical Information Science, pp. 1–23. Diakité, Abdoulaye Abou and Sisi Zlatanova (2016b). “Valid Space Description in BIM for 3D Indoor Navigation”. In: International Journal of 3-D Information Modeling (IJ3DIM) 5.3, pp. 1–17. Du, Xiaoyu, Regina Bokel, and Andy van den Dobbelsteen (2014). “Building microclimate and summer thermal comfort in free-running buildings with diverse spaces: A Chinese vernacular house case”. In: Building and Environment 82, pp. 215– 227. Fazio, P et al. (2007). “IFC-based framework for evaluating total performance of building envelopes”. In: Journal of architectural engineering 13.1, pp. 44–53. Fichtner, Florian W (2016). “Semantic enrichment of a point cloud based on an octree for multi-storey path-finding”. MA thesis. TUDelft. Gaisbauer, Christian and Andrew U Frank (2008). “Wayfinding model for pedestrian navigation”. In: AGILE 2008 Conference-Taking Geo-information Science One Step Further, University of Girona, Spain. Vol. 9. Galˇcík, František and Miroslav Opiela (2016). “Grid-based indoor localization using smartphones”. In: Indoor Positioning and Indoor Navigation (IPIN), 2016 International Conference on. IEEE, pp. 1–8. Goshayeshi, Danial et al. (2013). “A review of researches about human thermal comfort in semi-outdoor spaces”. In: European Online Journal of Natural and Social Sciences 2.4, p. 516. Gröger, Gerhard et al. (2008). “OpenGIS city geography markup language (CityGML) encoding standard”. In: Open Geospatial Consortium Inc, pp. 1–234. He, Jiang and Akira Hoyano (2010). “Measurement and evaluation of the summer microclimate in the semi-enclosed space under a membrane structure”. In: Building and Environment 45.1, pp. 230–242. Hilsenbeck, Sebastian et al. (2014). “Graph-based data fusion of pedometer and WiFi measurements for mobile indoor positioning”. In: Proceedings of the 2014 ACM International Joint Conference on Pervasive and Ubiquitous Computing. ACM, pp. 147– 158. Hooff, Twan Antonius Johannes van and Bert Blocken (2009). “Computational analysis of natural ventilation in a large semi-enclosed stadium”. In: Proceedings of the 5th European and African Conference on Wind Engineering. Firenze University Press., pp. 1–11. Hoogendoorn, Serge P and Piet HL Bovy (2004). “Pedestrian route-choice and activity scheduling theory and models”. In: Transportation Research Part B: Methodological 38.2, pp. 169–190. Hughes, Roger L (2002). “A continuum theory for the flow of pedestrians”. In: Transportation Research Part B: Methodological 36.6, pp. 507–535.

REFERENCE

51

Hwang, Ruey-Lung and Tzu-Ping Lin (2007). “Thermal comfort requirements for occupants of semi-outdoor and outdoor environments in hot-humid regions”. In: Architectural Science Review 50.4, pp. 357–364. Indraganti, Madhavi (2010). “Adaptive use of natural ventilation for thermal comfort in Indian apartments”. In: Building and environment 45.6, pp. 1490–1507. Kallmann, Marcelo and Mubbasir Kapadia (2014). “Navigation meshes and realtime dynamic planning for virtual worlds”. In: ACM SIGGRAPH 2014 Courses. ACM, p. 3. Khan, Aftab Ahmed, Zhihang Yao, and Thomas H Kolbe (2015). “Context Aware Indoor Route Planning Using Semantic 3D Building Models with Cloud Computing”. In: 3D Geoinformation Science. Springer, pp. 175–192. Kim, Jiyoeng, Taeyeon Kim, and Seung-Bok Leigh (2011). “Double window system with ventilation slits to prevent window surface condensation in residential buildings”. In: Energy and Buildings 43.11, pp. 3120–3130. Kim, Kwangho, Sanghyun Park, and Byungseon Sean Kim (2008). “Survey and numerical effect analyses of the market structure and arcade form on the indoor environment of enclosed-arcade markets during summer”. In: Solar Energy 82.10, pp. 940–955. Kim, Taeyeon, Shinsuke Kato, and Shuzo Murakami (2001). “Indoor cooling/heating load analysis based on coupled simulation of convection, radiation and HVAC control”. In: Building and Environment 36.7, pp. 901–908. Koide, Shohei and Masami Kato (2005). “3-d human navigation system considering various transition preferences”. In: Systems, Man and Cybernetics, 2005 IEEE International Conference on. Vol. 1. IEEE, pp. 859–864. Köster, Gerta, Daniel Lehmberg, and Felix Dietrich (2016). “Is Slowing Down Enough to Model Movement on Stairs?” In: Traffic and Granular Flow’15. Springer, pp. 35– 42. Kourogi, Masakatsu et al. (2006). “Indoor/outdoor pedestrian navigation with an embedded GPS/RFID/self-contained sensor system”. In: Advances in Artificial Reality and Tele-Existence, pp. 1310–1321. ¯ ˙ Marija (2014). “Space Subdivision for Indoor Navigation”. MA thesis. Kruminait e, Dissertação (Mestrado em Geomática). Delft University of Technology. ¯ ˙ Marija and Sisi Zlatanova (2014). “Indoor space subdivision for indoor Kruminait e, navigation”. In: Proceedings of the Sixth ACM SIGSPATIAL International Workshop on Indoor Spatial Awareness. ACM, pp. 25–31. Lee, J et al. (2014). “OGC IndoorGML”. In: Open Geospatial Consortium standard. Li, Ki-Joune (2008). “Indoor space: A new notion of space”. In: International Symposium on Web and Wireless Geographical Information Systems. Springer, pp. 1–3. — (2016). “IndoorGML-A Standard for Indoor Spatial Modeling”. In: International Archives of the Photogrammetry, Remote Sensing & Spatial Information Sciences 41, pp. 701–704. Li, Shaogang (1994). “Users’ behaviour of small urban spaces in winter and marginal seasons”. In: Architecture and Behaviour 10.1, pp. 95–109. Lin, Tzu-Ping (2009). “Thermal perception, adaptation and attendance in a public square in hot and humid regions”. In: Building and environment 44.10, pp. 2017– 2026. Lin, Tzu-ping, Andreas Matzarakis, and Jia-jian Huang (2006). “Thermal comfort and passive design strategy of bus shelters”. In: Lin, Tzu-Ping et al. (2008). “The comparison of thermal sensation and acceptable range for outdoor occupants between Mediterranean and subtropical climates”. In: Proceedings 18th International Congress on Biometeorology.

52

REFERENCE

Liu, Liu and Sisi Zlatanova (2011). “Towards a 3D network model for indoor navigation”. In: Urban and Regional Data Management, UDMS Annual, pp. 77–92. — (2013). “A two-level path-finding strategy for indoor navigation”. In: Intelligent systems for crisis management. Springer, pp. 31–42. — (2015). “An Approach for Indoor Path Computation among Obstacles that Considers User Dimension”. In: ISPRS International Journal of Geo-Information 4.4, pp. 2821–2841. Liu, Xiao Hu et al. (2013). “Urban Solar Updraft Tower Integrated with Hi-Rise Building–Case Study of Wuhan New Energy Institute Headquarter”. In: Applied Mechanics and Materials. Vol. 283. Trans Tech Publ, pp. 67–71. Mandloi, Deelesh and Jean-Claude Thill (2010). “Object-oriented data modeling of an indoor/outdoor urban transportation network and route planning analysis”. In: Geospatial Analysis and Modelling of Urban Structure and Dynamics. Springer, pp. 197–220. May, Andrew J et al. (2003). “Pedestrian navigation aids: information requirements and design implications”. In: Personal and Ubiquitous Computing 7.6, pp. 331–338. Melson, Gail F (1977). “Sex differences in use of indoor space by preschool children”. In: Perceptual and Motor Skills 44.1, pp. 207–213. Monteiro, Leonardo M and Marcia P Alucci (2007). “Transitional spaces in São Paulo, Brazil: mathematical modeling and empirical calibration for thermal comfort assessment”. In: Building Simulation. Mortari, Filippo et al. (2014). “" Improved geometric network model"(IGNM): A novel approach for deriving connectivity graphs for indoor navigation”. In: ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences 2.4, p. 45. Munkres, James R (1984). Elements of algebraic topology. Vol. 2. Addison-Wesley Menlo Park. Nagel, Claus et al. (2010). “Requirements and Space-Event Modeling for Indoor Navigation - How to simultaneously address route planning, multiple localization methods, navigation contexts, and different locomotion types”. In: Open Geospatial Consortium, pp. 1–54. Nasir, Nazliah Hani Mohd, Farha Salim, and Maheran Yaman (2014). “THE POTENTIAL OF OUTDOOR SPACE UTILIZATION FOR LEARNING INTERACTION”. In: UMRAN2014, Fostering Ecosphere in Built Environment, International Islamic University Malaysia, pp. 1–17. Neufert, Ernst, Peter Neufert, and Johannes Kister (2012). Architects’ data. John Wiley & Sons. Nikander, Jussi et al. (2013). “Indoor and outdoor mobile navigation by using a combination of floor plans and street maps”. In: Progress in Location-Based Services. Springer, pp. 233–249. Nourian, Pirouz et al. (2016). “Voxelization algorithms for geospatial applications: Computational methods for voxelating spatial datasets of 3D city models containing 3D surface, curve and point data models”. In: MethodsX 3, pp. 69–86. Ortiz, Alberto et al. (2015). “Saliency-driven Visual Inspection of Vessels by means of a Multirotor”. In: The Workshop on Vision-Based Control & Navigation of Small, pp. 20–46. Pagliarini, G and S Rainieri (2011a). “Dynamic thermal simulation of a glass-covered semi-outdoor space with roof evaporative cooling”. In: Energy and Buildings 43.2, pp. 592–598.

REFERENCE

53

— (2011b). “Thermal environment characterisation of a glass-covered semi-outdoor space subjected to natural climate mitigation”. In: Energy and Buildings 43.7, pp. 1609– 1617. Philokyprou, Maria et al. (2017). “Thermal performance assessment of vernacular residential semi-open spaces in Mediterranean climate”. In: Indoor and Built Environment, p. 1420326X17699037. Richter, Kai-Florian, Stephan Winter, and Sigit Santosa (2011). “Hierarchical representations of indoor spaces”. In: Environment and Planning B: Planning and Design 38.6, pp. 1052–1070. Rodenberg, OBPM, E Verbree, and S Zlatanova (2016). “Indoor A* Pathfinding Through an Octree Representation of a Point Cloud”. In: ISPRS Annals of Photogrammetry, Remote Sensing and Spatial Information Sciences, pp. 249–255. Rusné, Šileryté, Cavic Ljiljana, and Jose Nuno Beirao (2017). “Automated generation of versatile data model for analyzing urban architectural void”. In: Computers, Environment and Urban Systems 66, pp. 130–114. Saeedi, Sara (2013). “Context-Aware Personal Navigation Services Using Multi-level Sensor Fusion Algorithms”. PhD thesis. UNIVERSITY OF CALGARY. Sevtsuk, Andres and Michael Mekonnen (2012). “Urban network analysis”. In: Revue internationale de géomatique–n 287, p. 305. Shelke, SN et al. (2016). “Autonomous Campus Driver Assistant for Indoor and Outdoor Navigation”. In: Imperial Journal of Interdisciplinary Research 2.7. Slingsby, Aidan and Jonathan Raper (2008). “Navigable space in 3D city models for pedestrians”. In: Advances in 3D geoinformation systems, pp. 49–64. Spagnolo, Jennifer and Richard De Dear (2003). “A field study of thermal comfort in outdoor and semi-outdoor environments in subtropical Sydney Australia”. In: Building and environment 38.5, pp. 721–738. Stoffel, Edgar-Philipp, Bernhard Lorenz, and Hans Jürgen Ohlbach (2007). “Towards a semantic spatial model for pedestrian indoor navigation”. In: ER Workshops. Springer, pp. 328–337. Teo, Tee-Ann and Kuan-Hsun Cho (2016). “BIM-oriented indoor network model for indoor and outdoor combined route planning”. In: Advanced Engineering Informatics 30.3, pp. 268–282. Turrin, Michela et al. (2009). “Digital design exploration of structural morphologies integrating adaptable modules”. In: A design process based on parametric modeling. in: Proceedings of Caad Futures 2009 International Conference. Joining languages, cultures and visions. Montreal, Canada, pp. 17–19. Turrin, Michela et al. (2012). “Performative skins for passive climatic comfort: A parametric design process”. In: Automation in Construction 22, pp. 36–50. Tyson, Glenn M, Richard M Schuler, and Michael P Leonhardt (2012). In-grade lighting system. US Patent 8,313,208. Van Timmeren, A and M Turrin (2009). “Case study ?the Vela Roof–UNIPOL?, Bologna: use of on-site climate and energy resources”. In: WIT Transactions on Ecology and the Environment 121. Wang, Weiping et al. (2016). “Indoor-Outdoor Detection Using a Smart Phone Sensor”. In: Sensors 16.10, p. 1563. Winter, Stephan (2012). “Indoor spatial information”. In: International Journal of 3-D Information Modeling (IJ3DIM) 1.1, pp. 25–42. Worboys, Michael (2011). “Modeling indoor space”. In: Proceedings of the 3rd ACM SIGSPATIAL International Workshop on Indoor Spatial Awareness. ACM, pp. 1–6. Wu, Hua et al. (2007). “Applying HTA method to the design of context-aware indoor navigation for the visually-impaired”. In: Proceedings of the 4th international

54

REFERENCE

conference on mobile technology, applications, and systems and the 1st international symposium on Computer human interaction in mobile technology. ACM, pp. 632–635. Xu, Man, Shuangfeng Wei, and Sisi Zlatanova (2016). “AN INDOOR NAVIGATION APPROACH CONSIDERING OBSTACLES AND SPACE SUBDIVISION OF 2D PLAN.” In: International Archives of the Photogrammetry, Remote Sensing & Spatial Information Sciences 41. Yan, Jinjin et al. (2016). “Indoor Spatial Structure and Mapping Methods for Realtime Localization”. In: Geomatics and Information Science of Wuhan University 41.8, pp. 1079–1086. Yang, Liping and Michael Worboys (2011a). “A navigation ontology for outdoorindoor space:(work-in-progress)”. In: Proceedings of the 3rd ACM SIGSPATIAL international workshop on indoor spatial awareness. ACM, pp. 31–34. — (2011b). “Similarities and differences between outdoor and indoor space from the perspective of navigation”. In: Poster presented at COSIT. Yang, Wei, Nyuk Hien Wong, and Steve Kardinal Jusuf (2013). “Thermal comfort in outdoor urban spaces in Singapore”. In: Building and Environment 59, pp. 426– 435. Yuan, Wenjie and Markus Schneider (2010). “Supporting 3D route planning in indoor space based on the LEGO representation”. In: Proceedings of the 2nd ACM SIGSPATIAL International Workshop on Indoor Spatial Awareness. ACM, pp. 16–23. Zhou, Pengfei et al. (2012). “Iodetector: A generic service for indoor outdoor detection”. In: Proceedings of the 10th acm conference on embedded network sensor systems. ACM, pp. 113–126. Ziemer, Verena, Armin Seyfried, and Andreas Schadschneider (2016). “Congestion dynamics in pedestrian single-file motion”. In: Traffic and Granular Flow’15. Springer, pp. 89–96. Zlatanova, Sisi, Liu Liu, and George Sithole (2013). “A conceptual framework of space subdivision for indoor navigation”. In: Proceedings of the Fifth ACM SIGSPATIAL International Workshop on Indoor Spatial Awareness. ACM, pp. 37–41. Zlatanova, Sisi et al. (2014). “Space subdivision for indoor applications”. In: GISt Report No. 66.