Autonomous Ultrasonic Indoor Tracking System

2008 International Symposium on Parallel and Distributed Processing with Applications

Autonomous Ultrasonic Indoor Tracking System Junhui Zhao, Yongcai Wang NEC Labs, Beijing, China {zhaojunhui,wangyongcai}@research.nec.com.cn Abstract

to provide location based services. The typical examples are Bat [9] and Cricket [10]. However, these systems remain in the laboratory or university and are hard to put into real use. The main challenge is the considerable installation and calibration effort, which is special requirement for a location system before being put into use. To bootstrap a conventional ultrasonic location system, firstly networked reference points are required to be deployed and distributed into practical environment, which poses considerable cost for system installation. Then, the positions of the reference points have to be measured or calibrated to form a reference space, which also needs a lot of manual efforts. Thereafter, the position of the object can be calculated with respect to the reference points.

This paper proposes the Autonomous Ultrasonic Indoor Tracking System (AUITS), an ultrasound based system for locating and tracking mobile objects inside a building. Ultrasound shows promise to be exploited for a practical indoor location system due to its high accuracy ranging, low cost, safety, and imperceptibility. However, conventional ultrasonic location systems pose such challenges as high installation cost and manual calibration. The key idea of AUITS is to use only one autonomous device, Positioning on One Device (POD), to not only process signal acquisition but also conduct position computation. Structural topology is designed to make POD easily deployed and easily calibrated. In addition, a structural localization algorithm is proposed to provide an effective and affordable algorithm for POD to calculate the object’s position. We describe the hardware prototype implementation of AUITS and evaluate its performance both experimentally and with simulation. The results show that the coverage area of a POD can reach 65 m2 and the positioning error is less than 15 cm with over 90% probability.

To address these limitations in current approaches, we propose AUITS, an autonomous, ultrasound based system for locating mobile objects inside a building. The key idea of AUITS is Positioning on One Device (POD), which uses one positioning device to conduct both signal acquisition and position computation. AUITS is composed by POD and Tags. A Tag containing both a RF transceiver and an US transmitter is carried by the mobile object to be located that works in active transmission mode. A POD integrates single RF transceiver and multiple US receivers into one device so that it can be easily deployed into application environment and autonomously localizes positions of the mobile Tags. Structural topology is designed for POD to reduce the installation and calibration efforts, especially, enabling automatic self-calibration. Based on structural topology, structural localization algorithm is presented to improve the efficiency of position estimation process on POD. The design of AUITS offers several advantages; easy installation, easy calibration and highly accurate location. We implement the hardware prototype of AUITS from off-the-shelf components, including POD and Tag. The performance of AUITS is evaluated with simulation and experiment.

1 Introduction Location information can be utilized to extract the geographical relationship between users and the environments to further understand user behavior [1]. The importance and promise of location-aware applications has led to the design and implementation of systems for providing location information, particularly in indoor and urban environments [2, 3]. In general, there are various kinds of signals commonly used for indoor location systems, including Infrared [4], radio frequency (RF) [5-8], Ultrasound (US) [9-12], audible sound [13], etc. Compared with other signals, we favor that ultrasound is a promising means to be exploited for a practical indoor location system due to its high accuracy ranging, low cost, safety and imperceptibility by user. So far, there are already some ultrasonic indoor location systems 978-0-7695-3471-8/08 $25.00 © 2008 IEEE DOI 10.1109/ISPA.2008.37

The rest of this paper is organized as: we present the system overview in Section 2. The design philosophy is introduced in Section 3. The prototype implementation and performance evaluation are presented in Section 4 and 5, respectively. Finally, Section 6 concludes the paper. 532

RF

POD

US

Tag

(1) Emission

(2) Synchronization

Context Information Server

(3) Acquisition

(4) Calculation

Figure 1. Work Flow of AUITS

2 System Overview

from POD via wire or wireless network to a context information server.

Our aim is to develop an accessible indoor ultrasonic location system , which is easily deployed, auto-calibrated, highly accurate, so that the real-time position and movement trajectories of the mobile object can be tracked. In AUITS, a tag is carried on a mobile object to be located, which works in active transmission mode. An autonomous positioning device, POD, is mounted on the ceiling of the monitoring area to carry out airborne US signal collection and position computation. The tag may contain a RF transceiver and a narrowband US transmitter that is low cost. POD contains single RF transceiver and multiple US receivers. The work flow of AUITS is shown in Fig.1.

3 POD Design 3.1

Structural Topology

Installation and calibration efforts are important factors to make a location system feasible for practical applications. One remarkable benefit of the proposed AUITS is the structural property of POD, which greatly reduces the installation and calibration efforts, especially, enabling the structural-based self-calibration. Multiple US receivers are integrated into the single device, POD, and these receivers are uniformly distributed in POD so that structural topology is constructed. Fig.2 illustrates some typical structure of our proposed POD.

1. Emission Mobile tag sends an RF signal and ultrasonic pulses simultaneously. 2. Synchronization Hearing the RF signal, POD synchronizes all its connected US receivers to wait for the succeeding US.

3

1 l

3. Acquisition US receiver detects the airborne US signal emitted from mobile tag and reports the detecting time to POD. Then POD calculates distances from US receiver to the tag based on the Time of Arrivals (TOA).

2

0

X axis

(x0,y0)0 4 3

(a) Square Style

4. Calculation A structural localization algorithm is utilized for position estimation. Position results are sent

4 0

0 5

2 X axis

l

1

(x0,y0) 6

(b) Hexagon Style

Figure 2. Typical Structure of POD

533

Basically, the proposed POD appearances in above figure are central symmetric, where one US receiver is located at POD center and others are uniformly distributed along the POD edges. In particular, the surrounding receivers can form a regular shape, such as a square or a hexagon.

central US receiver as the origin (x0 = 0; y0 = 0; z0 = 0). This does not involve any manual calibration. After position computation of POD, a relative position of the object will be sent to a context server for location based services. If an absolute position coordinate of the object is required, only the POD center’s position needs to be determined. The measurement of the POD center absolute coordinate requires less manual effort compared with conventional ultrasonic location systems. The absolute coordinate of the object can be easily mapped from the POD reference space to the absolute coordinate space. Based on the above analysis, it is evident that structural topology can eliminate the efforts of measuring the distances and angles among US receivers, making POD more convenient and accessible for practical application scenarios.

3.1.1 Easy Installation Fig.3 shows the POD installation process. An initial form of POD is just like a compact ”Frisbee” that contains single RF transceiver and multiple US receivers. While in use, the surrounding receivers can be extended to the outside so that POD has an umbrella-like appearance. Instead of deploying one by one networked sensors in the building as presented in prior art [9,10], POD can be easily installed only once. Therefore the installation efforts will be reduced tremendously. After installation, the coordinates of the US receivers are automatically obtained according to structural topology design. Surrounding

Central

3.2

Structural Localization

The block diagram of our proposed structural localization algorithm is shown in Fig.4. The inputs are the simultaneous TOA measurements from different US receivers of POD. A TOA reliability filter is used for pre-processing in order to reject the outliers from the TOA measurements. After that, a unique estimated position can be obtained based on structural trilateration.

POD (Spread)

Ceiling Installation

TOA Detections

POD (Compact)

TOA Reliability Filter

Structural Trilateration

Unique Position Estimation

Figure 4. Block Diagram of Structural Localization

Figure 3. POD Installation Process

3.1.2 Easy Calibration

3.2.1 The TOA Reliability Filter

In calibration phase, rather than manually measure the coordinates of each US receiver individually, POD can easily infer the coordinates of all receivers based on the structural topology. To be specific, as the direction of the first US receiver is set as the X axis, coordinates of all the other receivers can be obtained as follows. In counterclockwise direction, the coordinates of the ith surrounding receiver is given by:  ³ ´  xi = l · cos 2π·(i−1) ³ n−1 ´ i = 1, ..., n (1)  yi = l · sin 2π·(i−1) n−1

Ultrasound signal may be reflected by obstacles and such reflected signal will lead to unexpected localization error that should be taken as outlier to be removed from our localization algorithm. We propose a Fast Outlier Rejection (FOR) algorithm to identify and eliminate these outliers from the simultaneous distance estimations. The basic idea is that the difference of two transmitter-receiver distances can not be larger than the distance between the receivers. In the algorithm, we assume the minimum survived TOA measurement is the most reliable. Based on the assumption, the distances are sorted in ascending order. The maximum and the minimum distances are checked according to triangle inequalities. ½ di − dj < Ei,j (2) di + dj > Ei,j

where n is the number of all US receivers and l is the distance between the surrounding receiver and POD center. It should be noted that in the position computation phase, POD always calculates the object’s position relative to itself. That is, POD automatically sets the coordinate of the

where di and dj are the maximum and minimum distance measurements, respectively, and Ei,j is the distance be-

534

tween receiver i and receiver j. If both inequalities are satisfied, the largest distance is regarded as a direct-path signal and a reliable TOA measurement. The filtering process will end and all TOA distance measurements will be accepted as reliable values. Otherwise, the largest distance is classified as an outlier since there is high probability of it being a reflected signal. The second maximum distance measurement is then checked. Such checking process repeats until this condition is satisfied.

3

0

4

3

2

5

1

3.2.2 Structural Trilateration POD employs trilateration algorithm to infer the object’s position from three TOA measurements. When more than three reliable TOA measurements are obtained after reliability filtering, the conventional localization procedure is multilateration that uses all reliable TOA for position calculation. However, due to multilateration’s high computational cost, this is not feasible to POD. In our design, a structural trilateration method is proposed to utilize only three of all TOA samples to get an unique position estimation. Structural trilateration is based on TOA refinement. According to [14], the geometric relation of reference points can significantly affect the localization result. The farther separated the reference points are, the better the localization result will be. Conversely, if the three reference points almost lie in a line, there will be a high probability of flip ambiguity in the localization result. Thus, we select the three TOA samples based on a ”Max Separation” criteria, that is, the distribution between these reference points should be as dispersed as possible. Since POD has the structural topology, we can easily define the separation degree according to the possible Three-Point distributions on POD. For example, in a hexagonal POD, there are four potential Three-Point distributions for trilateration. These are shown in Fig.5. In this figure, US receivers on POD are indexed and the central receiver is set as 0. In (a) large equilateral triangles can be formed by joining US receivers (1,3,5) or (2,4,6). In (b), there are 12 possible right angled triangles because each edge has two other points to construct a right angled triangle. For example, (1,2,4) and (1,3,4). In (c), since each edge can determine one small equilateral triangle, there are 6 possible small equilateral triangles. In (d), the center receiver corresponds to 6 possible obtuse angle triangles. For example, (0,2,6) and (0,1,3). Also, each surrounding receiver can form an obtuse angle triangle, such as (1,2,6). Therefore, there are 12 possible obtuse angle triangles. It should be pointed out here that the separation degree can be ordered by (a)>(b)>(c)>(d). That is, the ThreePoint distributions in (a) have maximum separation degree so that three TOA samples with large equilateral triangles

5

1

5

6

6

(b) right angle triangle 3

2

0

4

0

4

(a) large equilateral triangle 3

2

1

0

4

6

(c) small equilateral triangle

2

5

1

6

(d) obtuse angle triangle

Figure 5. Separation Degree based on ThreePoint Distribution for Hexagon-style POD

should have the highest trilateration selection priority. On the other hand, (d) represents the minimum separation degree but can still be used for position calculation. Any Three-Point distributions except the above, such as (1,0,4), will be abandoned because no position result can be obtained by them.

4 Prototype Implementation 4.1

POD

POD is implemented using inexpensive, off-the-shelf, simple hardware parts. Basic parameters for POD prototype design can be found in Table 1. It includes a microcontroller (such as ATMEL 128 processor), a RF Transceiver for time synchronization and object ID recognition and a set of US Receivers. A CPLD (Complex Programmable Logic Devices) is used to setup a single clock for all US receivers. As soon as the synchronization time is detected by microcontroller, CPLD is signaled for timing all US receivers so that these receivers can be synchronized with minimal error. Since the proposed structural localization needs low computation cost, it can be performed well on the microcontroller. The arrangement of US receivers has one placed in the POD center and others at surrounding edges. For example, a POD board can be implemented as a square for 5 US receivers and a hexagon for 7 US receivers. Around the POD board, each US receiver is connected via a telescopic rod. The US receiver can be pulled outside to form an umbrella-like topology. Fig.6 gives the implementation of the Hexagon-shape POD, the typical size parameters for

535

5.1

such a POD is that the distance between center and surrounding US receiver is 23 cm in Compact Mode and the distance is 60 cm in Spread Mode.

Coverage

In AUITS, coverage area becomes an important parameter of POD, indicating its suitable application scope. Basically, coverage area of POD is mainly affected by three parameters: • Number of surrounding US receivers (SU) • Spread distance (SD) • Tilt angle of surrounding US receivers (TA) Here, The spread distance means the distance from surrounding receiver to POD center. The tilt angle means the offset degree when surrounding receiver is rotated towards outside. The evaluation for coverage property is carried out in a designed simulation platform based on OMNeT++ [15]. The simulation platform is illustrated in Fig.7. We set up a simulation environment where one POD is mounted on the ceiling and a mobile Tag moves around in a room, which is 16m wide and 18m long. The POD height is 2.5m to ground and the Tag height is 0.9m to ground. In simulation, the mobile Tag moves on grids of a horizontal plane and we set the grid size to 20cm∗20cm. In each grid, it broadcast RF+US signals for localization. According to structural localization, at least three reliable TOA measurements are necessary for accurate positioning. Therefore, if in one grid at least three TOA are detected by POD, such a grid can be counted into the covered area of POD. By counting the number of grids covered by POD, we can achieve the coverage area of POD.

Figure 6. Prototype for Hexagon-style POD

Table 1. Parameters of POD hardware prototype Name RF Chip RF frequency RF data rate RF transmission distance US Transducer US frequency US propagation distance

4.2

Value CC1000 1 433MHz 19.2Kbps 30m-60m 255-400SR12/ST12 2 40KHz 10 meters

Tag

Similar with POD prototype, tag is also implemented using low cost, off-the-shelf and simple hardware components. It contains a microcontroller, a RF transceiver and an US transmitter. Basically, the tag is implemented for goals of small size, low power consumption and user friendly. Two kinds of tag are manufactured. One is pen-style for person tracking and the other is small box-style for asset tracking.

5 Simulation and Experiment Figure 7. Simulation in OMNet++

We performed a number of simulations and experiments to evaluate AUITS performance, mainly on coverage, accuracy and robustness. 1 Please 2 Please

In our simulation, SD varies from 40 cm to 100 cm; TA varies from 0 to 15 degree; and SU varies as 4, 6, 8. The coverage area of One POD is statistically obtained in Table.2.

refer to http://www.chipcon.com/ refer to http://www.mouser.com/kobitone/

536

Table 2. Coverage area of POD according to variant parameter settings SD(cm) TA = 0o TA= 15o SU = 4 SU = 6 SU = 8 SU = 4 SU = 6 SU = 8 40 36.6 38.3 40.4 41.3 47.3 56.8 60 37.4 40.9 43.8 42.7 51.1 58.2 80 38.2 42.3 46.2 43.5 54.1 62.8 100 39 44.5 50.8 44.2 56 65.9 The first point to note is that basically coverage of POD can be increased by surrounding US receivers, tilt angle and spread distance. The overall coverage area of POD varies from 36.6 m2 to 65.9m2 in different parameter setting. Especially, the largest coverage can be achieved at the case of SD = 100cm, TA = 15o and SU = 8. Secondly, coverage area increases slightly with spread distance. For example, in case of SU=6 and TA=0o , the coverage areas change relatively smaller when SD ranges from 40 cm to 100 cm. More exactly, only an increase of 6.2 m2 can be obtained when spread distance is increased. Thirdly, significant improvement can be achieved when the US receiver is rotated toward outside. For example, in case of SD=60cm and SU=6, the coverage area is 51.1m2 when TA=15o , yielding an improvement of 21.7% compared with TA=0o . These results imply that POD can be designed more compact while preserving its coverage so that the accessibility of POD will be improved. To further improve POD prototype, we only need to shorten spread distance of POD and enlarge tilt angle accordingly.

5.2

Wall Corridor

15

11

10 9 2

8

4

POD 16 17 18 19

3

20 21

6

1

22 23 24 25 26 27 5 13

7

12

14

Test points near POD center Test points away from POD center Test Points far from POD center Test Points in error distribution evaluation

Figure 8. Test Points in Stationary Evaluation

Fig.9 gives the results of stationary position experiment. From the plot, we can firstly observe that AUITS system can yield highly accurate positioning results when the object is below and near POD center. The positioning error is less than 10cm with over 90% probability. Secondly, when the object is not directly below POD center, AUITS system also yields highly accurate positioning results, leading to 15cm positioning error with over 90% probability. Finally, when the object is far from POD center, the performance of accuracy declines sharply, because less US receivers can detect the US emitted from the transmitter far from POD. But the overall accuracy performance is still acceptable for most application cases.

Accuracy

Experiments are carried out in our Labs. The room is 5 meters long and 3 meters wide, surrounded by walls and there is a corridor beside the room. There are tables, chairs and a whiteboard in this room. A POD with a RF transceiver and 7 US receivers is mounted on the ceiling of this room. The height from POD to ground is 2.6m. The positions of POD in the plan view of the room are shown in Fig. 8. 5.2.1 Stationary Test In stationary test, we choose three categories of test points for accuracy evaluation.

5.2.2 Error Distribution

• Test points near the center of POD are indexed by {15} marked as small squares in Fig.8;

We further evaluate positioning error distribution affected by distance and angle. Test points are selected to form a line below POD that are indicated by cross point as illustrated in Fig.8. The error distribution result is shown in Fig.10. The origin point of X-axis is the projection of POD center on the surface of the object and the X-axis represents the distance from object to the projection center of POD.

• Test points away from the center of POD are {6-10} marked as small rectangles; • Test point far from the center of POD are {11-15} marked as circles.

537

1

30

0.8 10 y(cm)

0.7 near POD center away from POD center far from POD center

0.6 0.5

0 -10

0.4

-20

0.3

-30

0.2 0.1 0 0

5

10

15 20 25 30 Positioning Error (cm)

35

40

-30 -20 -10 20 10 0 -10 -20

0 10 x(cm)

1

20

30

2 3 Experimental Tracks

4

Figure 11. Position Error in Dynamic Test

Figure 9. Cumulative Distribution of Position Error for Stationary Test

5.3 20

Robustness

We also design test cases for evaluating the robustness of the system. In particular, we focus on investigating the influence of people moving to the system because mobility is quite common and may block or reflect the US signals. In that case, it is a very severe threat to current ultrasonic indoor location systems. Here, we also evaluate the case of Stationary Test and Dynamic Test. In order to investigate robustness, when the tag broadcasts US, a person keeps moving around the tag, and blocked the tag with hand or arm occasionally.

18 16 Positioning error(cm)

True track Experimental track 1 Experimental track 2 Experimental track 3 Experimental track 4

20

Positioning error (cm)

cumulative probability

0.9

14 12 10 8 6 4 2 0 -320-260 -200 -140 -80 -20 40 100 160 220 280 Distance from POD center(cm)

5.3.1 Stationary Test Figure 10. Position Error According to Distance From Transmitter to POD Center

1

cumulative probability

0.9

According to the result, positioning error increases slowly as the distance increases from zero to two meters, yet increases rapidly when the distance is larger than two meters. The overall positioning error is less than 15cm, indicating a good accuracy performance of AUITS.

0.8 0.7 0.6 0.5 0.4

Near POD Center with noise Away from POD center with noise

0.3 0.2 0.1

5.2.3 Dynamic Test

0 0

Dynamic tests evaluate the system’s refresh rates and positioning accuracy as the object is moving. In this test, the US transmitter is put on the back of a swivel chair. It moves along the circular track as the chair is turned around. Four experiments are carried out. Fig.11 shows the comparison of the estimated tracks with the true circular track.The mean errors and standard derivations of the four experimental tracks are also plotted. The results show that as the transmitter moves dynamically, its position can be well tracked by the AUITS system. The overall mean error are less than ±10 cm.

5

10

15 20 25 30 Positioning Error (cm)

35

40

Figure 12. Position Error for Stationary Test in Noisy Environment For stationary test, we can see from Fig.12 that the accuracy performance degrades when noises are introduced. More exactly, when the object is near POD center, the positioning error is around 15 cm with 90% probability. As the object is away from POD center, the performance drops to less than 20 cm with 90% probability. The reason for

538

performance degradation is that reflections and blocks can cause errors for TOA measurement.

[2] J. Aspnes, T. Eren, D. K. Goldenberg, A. S.Morse, W.Whiteley, Y. R. Yang, B. D. O. Anderson, and P. N. Belhumeur, A Theory of Network Localization, IEEE Transactions onMobile Computing, vol. 5, no. 12, pp. 1663-1678, Dec. 2006.

5.3.2 Dynamic Test

[3] A. Asthana, M. Cravatts, and P. Krzyzanowski. An Indoor Wireless System for Personalized Shopping Assistance. In Proceedings of IEEE Workshop on Mobile Computing Systems and Applications, pages 69-74, Santa Cruz, California, December 1994. IEEE Computer Society Press.

True track Experimental track 1 Experimental track 2

30 20 y(cm)

10 0

[4] R. Want, etc, The Active Badge Location System, ACM Transactions on Information Systems (TOIS), Vol. 10 , Issue 1, Pages: 91 - 102, 1992.

-10 -20 -30 Positioning Error (cm)

-30 -20 -10 40 30 20 10 0 -10 -20

0 10 20 30 x(cm)

[5] P. Bahl and V. Padmanabhan. RADAR: An In-Building RFbased User Location and Tracking System. In Proc. IEEE INFOCOM (Tel-Aviv, Israel, Mar. 2000).

1 2 Experimental Tracks

[6] B. Kusy, A. Ledeczi and X. Koutsoukos, Tracking Mobile Nodes Using RF Doppler Shifts. in SenSys ’07: Proceedings of the 5th international conference on Embedded networked sensor systems. Sydney, Australia, ACM Press, pp. 29 - 42.

Figure 13. Position Error for Dynamic Test in Noisy Environment

[7] J. Hightower, C. Vakili, G. Borriello, and R. Want. Design and Calibration of the SpotON Ad-Hoc Location Sensing System. August 2001.

For dynamic test, two experimental tracks are tested and the estimated tracks are plotted in Fig.13. Comparing Fig.11 and 13, we can see the tracking performance becomes worse when noises are introduced. But the AUITS system can still provide a reasonable position estimation in this dynamic and noisy environment.

[8] L. M. Ni, Y. Liu, Y. C. Lau, and A. P. Patil. LANDMARC: Indoor Location Sensing Using Active RFID. In Proceedings of IEEE PerCom 2003, Dallas, TX, USA, March 2003. [9] A. Ward, A. Jones, and A. Hopper. A New Location Technique for the Active Office. In IEEE Personal Communications, Volume 4, no.5, October 1997.

6 Conclusion

[10] Nissanka B. Priyantha, A. Chakraborty, and H. Balakrishnan. The Cricket Location-Support System. In Proceedings of the Sixth International Conference on Mobile Computing and Networking (ACM MobiCom), Boston, Massachusetts, USA, August 2000.

In this paper, we designed, implemented and evaluated AUITS - an ultrasound-based system for locating and tracking mobile objects inside a building. AUITS demonstrates many desirable properties including ease of installation, auto-calibration and high accuracy. The design philosophy we followed is positioning on one device with structural topology, thus making the POD feasible and accessible as a practical location system. Besides, a structural localization algorithm is proposed to improve the positioning accuracy and reduce computational cost. In the future, there are several directions which might be pursued in order to further improve the ideas and algorithms presented in this paper, such as localization fusion of multiple POD for large application scenario and anti-collision protocol design for multiple tag tracking.

[11] C. Randell and H.L. Muller, ”Exploring the Dynamic Measurement of Position,” Proc.6th Int’l Symp. Wearable Computers (ISWC 02), IEEE CS Press, 2002, pp. 117-124. [12] S. Holm, O. Hovind, S. Rostad, and R. Holm, Indoors Data Communications Using Airborne Ultrasound. In Proceeding of ICASSP2005, Philadelphia, PA, USA, pp 957-960. [13] J. Scott and B. Dragovic. Audio Location: Accurate Lowcost Location Sensing. Pervasive 2005, pp. 1-18. [14] Z. Yang and Y. Liu, Quality of Trilateration: Confidence based Iterative Localization, in proceeding of IEEE ICDCS 2008, Pages: 446-453. [15] A. Varga, The OMNeT++ Discrete Event Simulation System, in Proceedings of the 15th European Simulation Multi conference (ESM’2001). European Council for Modeling and Simulation, Nottingham, UK, May 2001.

References [1] M. Weiser. The Computer for the 21st Century. Scientific American (September 1991)

539