SICE Annual Conference in Fukui, August 4-6, 2003 Fukui University, Japan
Evaluation of a new sensor system for ambulatory monitoring of human posture and walking speed using accelerometers and gyroscope K. Motoi, S. Tanaka, M. Nogawa, K. Yamakoshi Faculty of Engineering, Kanazawa University, 2-40-20 Kodatsuno, Kanazawa, Ishikawa 920-8667, Japan
[email protected] Abstract: Measurement of physical activity is one of the key subjects in the fields of rehabilitation and gerontology. From this point of view, we have been developing portable devices for monitoring human posture and walking speed in ambulatory subjects. In this paper, a new sensor system using three accelerometers and one gyroscope was developed, and its availability was evaluated. Keywords: Ambulatory monitoring, Human posture, Walking speed, Accelerometer, Gyroscope
1. Introduction Long-term ambulatory systems for monitoring cardio-vascular haemodynamic parameters, such as Holter type ECG and Ambulatory Blood Pressure Monitor (ABPM), have recently been commonly used not only in clinical medicine but also in basic research fields. It is also well known that these parameters are greatly influenced by several factors such as physical activity and posture changes. For the better understanding of the results obtained by these ambulatory cardiovascular monitoring devices, therefore, simultaneous measurement of the subject’s behavior and/or posture change during the monitoring period is indispensable. Importance of posture or activity monitoring is also well recognized in the fields of industrial health, rehabilitation, gerontology and so on. For example, in the field of gerontology, it is one of the key subjects for the elderly to maintain their activity in daily life in high condition to avoid becoming bedridden, and therefore, an objective measure of activity is essential. As indices of activity, the following measures are commonly used; e.g., movement or acceleration of body segment, frequency of posture change, walking speed, and so on. For these purposes, numerous numbers of instruments have been developed and some of those are commercially available. For example, the ActiGraph is a conventional wrist worn type activity monitor using an accelerometer and applied for the study of circadian rhythms, estimation of energy expenditure, assessment of rehabilitative programs, sleep research, and so on1). Recently, Aminian et al.(2002) reported an ambulatory physical activity monitor based on accelerometry with the sensor location of chest and thigh2). Sekine et al.(2002) proposed a method for discrimination of walking patterns from the signal of a tri-axial accelerometer fixed on a subject’s back in the lumbosacral region of the vertebral column3). These devices are available for rough estimation of physical activity or gait pattern, however, more detailed information is usually required especially in the fields of rehabilitation and gerontology. Taking the above matters into consideration, we have
been developing ambulatory instruments and reported about a portable device for monitoring human posture using miniature inclinometers4) and some results of application in gerontology field5). In this device, however, there still remains several practical drawbacks such as limited range of angle measurement, low accuracy in walking speed measurement due to inertial artifact during walking, higher maintenance requirements for the mechanical inclinometers, and so on. In order to overcome these problems, a new sensor system using accelerometer and gyroscope was designed and its accuracy in measuring human posture and walking speed were evaluated.
2. Outline of the system Fig. 1 shows the outline of the newly designed sensor system for monitoring human posture and walking speed. As shown in this figure, three accelerometers are attached to the subject’s trunk, thigh and shank. From the low frequency signal of these accelerometers, the angle to the gravitational direction of each part were measured, and thus, the human posture in sagittal plane could be easily discriminated. In the previous system4), range of the angle measurement was limited from 0 to 180 degree because of the characteristics of the inclinometer used. In the present system, however, as the two-axial accelerometers are used, this problem is solved and the angler range is expanded from 0 to 360 degrees. Method for estimating walking speed is basically similar to which reported by Miyazaki6). A gyroscope is attached to the thigh and angular change in saggital plane during walking is obtained by integrating the sensor signal. From these data and the subject’s leg length, step length is estimated by using a conventional gait model, and walking speed is calculated from the step period (see Fig. 2). In Fig. 3, photographs of the sensor units are also shown. These sensor units are attached to appropriate positions of the subject using specially designed sensor holders. The accelerometers used are tri-axial ones, however, for the purpose of the posture measurement, only two-axial outputs (x and z direction in Fig. 1) are used.
-563-
PR0001/03/0000-0563 ¥400 © 2003 SICE
3. Experiment
Z
Accelerometer X Z
Z
Accelerometer
X X Y
Gyroscope Z
Accelerometer X
Fig. 1 Outline of a new sensor system for measuring human posture and walking speed
T
Thigh angle(deg)
ǰ Time(sec)
Toe off
Heal contact
ǰ
ǰ
Toe off
L
Fig. 4 shows an experimental set up for evaluating accuracy of the system in measuring human posture and walking speed. The signals of each sensor were recorded on a digital disc analogue-signal/video recorder. During these experiments, sensor outputs were recorded with sampling frequency of 60Hz, and the subjects were simultaneously video recorded with frame speed of 30 frames per second using the recorder mentioned above. Stored signals of the accelerometers and the gyroscope were retrieved with appropriate filteration (accelerometer: DC~3Hz, gyroscope: 0.3~20Hz) and used for the angle and the angular velocity data calculation. In front of the video camera, three young healthy male subjects (aged from 21 to 23 yrs old) were asked to take several postures, e.g. sitting, standing, lying, etc. Four markers were attached on the appropriate positions of the subject (see Fig. 4). Actual angle to the gravitational direction was calculated using these markers. Estimated angle, i.e., angle calculated from output of the three accelerometers, was compared with actual angle for evaluating accuracy of human posture measurement. On the other hand, on the floor, appropriate number of black lines (length:30cm, interval:10cm) were marked using adhesive tape. Ten young healthy male subjects (aged from 21 to 23 yrs old) were asked to walk along the line makers with various walking speed including so called “small steppage gait” mimicking elderly. Actual walking speed is calculated using these lines (Step length¸Walking cycle). Estimated walking speed, which was obtained by the method mentioned above, was compared with actual speed for evaluating accuracy of walking speed measurement.
De Calculation of walking speed:
De = 4Lsin(ǰ/2), Ve = De/T
Fig. 2 Method for estimating walking speed by thigh angle measurement
٨Trunk and Shank sensor unit (Weight: 40.7g)
٤Digital disc recorder(InfoSync, ASTEC) Analogue signal: sampling frequency 60Hz Video signal: frame speed 30frame/s
٨Thigh sensor unit (Weight: 43.9g) 40mm
30mm
Sensor output 40mm
Marker
55mm
Accelerometer ٤Accelerometer Type: MC301, WACOH Size: 16167.3 mm Weight: 3.7 g Range: r1 g Frequency response: DC㨪15 Hz
Gyroscope
Digital video camera
Accelerometer
٤Gyroscope Type: ENC-03J, MURATA Size: 15.58.04.3 mm Weight: under 1.0 g Range: r300 deg/sec Frequency response: 50Hz
Marker(10cm intervals) Fig. 4 Experimental setup for evaluating accuracy of human posture and walking speed measurement
Fig. 3 Sensor units attached trunk, thigh and shank
-564-
4. Results and Discussion In Fig. 5, typical recordings of the angle to the gravitational direction of trunk, thigh and shank during three patterns (see (a), (b), (c) in Fig. 5) of posture change are shown. Definition of each angle (ǰ1,ǰ2 andǰ3) are schematically shown at the top of the figure. In these recordings, each line denotes the angle obtained from the filtered signal of the accelerometers, while the plots denote those obtained from VTR data (0.5sec intervals). As shown in these recordings, the angles measured by the accelerometers coincide well with those obtained from VTR (actual value) not only in the stable period, but also in the transient periods during posture changes. For more precise angle measurement during transient period (posture change), however, usage of gyroscope would be preferable. Fig. 6 shows the scatter diagrams between the angle obtained from the sensor output and actual values. Fairly good liner relationships are observed in each three part within wide range of the angle. It is expected that much higher accuracy in the angle measurement could be attained by introducing the acceleration data of the third axis (Y axis in Fig. 1). In Fig. 7 (a), typical recording of thigh angular velocity obtained from thigh gyroscope output is shown. In Fig. 7 (b), thigh angle change during walking which was obtained by integrating the signal from the gyroscope is shown. Regarding conventional gait model shown in Fig. 2, from the difference between maximum and minimum value of the thigh angle (ǰ), step length (D) was calculated using subject’s leg length (L). As the period of one step cycle (T) is known, and thus, the walking speed could be calculated from T and D. The method mentioned above is basically similar to which reported by Miyazaki, however, there is one big difference (or advantage) in the present system. It is well known that, in utilizing gyroscope for angle measurement, there are several problems such as difficulty in determining initial value of the angle, cumulative error by integration, and so on. In the present system, however, not only the gyroscope, but also the accelerometer are attached on the subject’s thigh, and therefore, “initial value of the thigh angle just before starting walk” could be easily determined (see the plot on the vertical axis of the recording in Fig. 7 b). And thus, the problem of the error accumulation by integration could be eliminated. Fig. 8 shows the correlation between the walking speed obtained from VTR and those from gyroscope. As shown in this figure, estimated values from the gyroscope coincide well with the actual values and quite good linear relationship is obtained within wide range of the walking speed. Marked differences between the subjects were not observed The gait model used here was a simple and conventional one (see Fig. 2), although, reasonable accuracy was attained as shown in Fig. 8. For the more precise measurement in walking speed or for the use of the patients such as under gait training, a new gait model considering the shank angle should be introduced.
ǰ1
ǰ2
(+)
ǰ3
(㧙)
(a) ǰ1 (deg)
120 90 60 30 0 -30 -60 -90 0
5
10
15
20
15
20
15
20
Time(sec)
ǰ2 (deg)
150 120 90 60 30 0 -30 0
5
10
Time(sec)
ǰ3 (deg)
120 90 60 30 0 -30 -60 0
5
10 Time(sec)
(b) ǰ1 (deg)
30 0 -30 -60 -90 -120 -150 -180 -210 0
5
10
15
20
15
20
15
20
15
20
15
20
15
20
Time(sec)
ǰ2 (deg)
120 90 60 30 0 -30 0
5
ǰ3 (deg)
10 Time(sec)
30 0 -30 -60 0
5
10 Time(sec)
(c) 60
ǰ1 (deg)
30 0 -30 -60 -90
ǰ2 (deg)
0
5
0
5
10 䌔䌩䌭䌥(sec)
120 90 60 30 0 -30 -60 -90 -120 10 䌔䌩䌭䌥(sec)
0 -30
ǰ3 (deg)
-60 -90 -120 0
5
10 䌔䌩䌭䌥(sec)
Fig. 5 Typical recordings of the angle to the gravitational direction of trunk, thigh and shank during three patterns of posture changes (a)(b)(c) and definition of angle at the top
-565-
120 90 60 30
1 2
0
3
-30 -60
y=0.99x+2.10 r=0.99
-90 -120 -120 -90
Estimated angle from accelerometer (deg)
㧔ǰ 㧕 2
-60
0
30
60
90
120
Actual angle from VTR (deg)
90 60 30
1
2
1.6
3 1.4
4
1.2
5
1.0
6
0.8
7
0.6
8
0.4
Ve=0.96Va+0.07
0.2
r=0.96
9 10
0.0
Actual walking speed from VTR: Va(m/s)
-60
y=0.98x+0.72 r=0.98
-90
-60 -30
0
30
60
90
Fig. 8 Correlation between the walking speed obtained from VTR and those from gyroscope including ten subjects
120
Actual angle from VT R (deg) 120 90
5. Conclusion
60 30
1 2
0
3
-30 -60
y=0.98x+0.36 r=0.98
-90
-60
-30
0
30
60
90
120
Actual angle from VTR (deg)
6 Scatter diagrams between the angle obtained from accelerometers output and actual values including three subjects (٤, ٌ, +)
angular velocity (deg/sec) (a) Thigh 200 150 100 50 0 -50 -100 -150
1
1.8
3
-30
-120 -120 -90
Fig.
2.0
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0
2
0
㧔ǰ 㧕 Estimated angle from accelerometer (deg)
-30
120
-120 -120 -90
3
Estimated speed from thigh gyroscope: Ve(m/s)
1
Estimated angle from accelerometer (deg)
㧔ǰ 㧕
From the results obtained in this study, it is clearly demonstrated that, using the newly designed sensor system, not only the human posture in saggital plane, but also the walking speed could be measured with reasonable accuracy. Especially in the lower range of walking speed with the value of less than 0.6m/s, which is usually within “normal range” of walking speed in elderly, fairly good results were obtained, indicating the applicability of the present sensor system for the gerontology fields. As the sensors used in this system are small and light weight, therefore, construction of an ambulatory system could be relatively easily attained. We have already developed a prototype device for ambulatory measurement using a portable data logger. Evaluation of this system is now proceeding, and the results will be soon reported elsewhere.
References 0
1
2
3
4 Time (sec)
㨀
angle (deg) (b) Thigh 40 30 20
ǰ
10 0 -10 -20
0
1
2
3
4 Time (sec) Initial angle obtained from the accelerometer output
Fig. 7 Typical recording of thigh angular velocity obtained from thigh gyroscope output (a) and thigh angle change during walking which was obtained by integrating the signal from the gyroscope (b)
1) http://www.mtiactigraph.com/ 2) K. Aminian, P. Robert, EE. Buchser, B. Rutschmann, D. Hayoz, M. Depairon: Physical activity monitoring based on accelerometry :validation and comparison with video observation, Med. Biol. Eng. Comput., 37-3, 304/308 (1999) 3) M. Sekine, T. Tamura, M. Akay, T. Fujimoto, T. Togawa, Y. Fukui: Discrimination of walking patterns using wavelet-based fractal analysis,IEEE Trans. Rehab. Eng.,10-3, 188/196 (2002) 4) S. Tanaka, K. Yamakoshi, P. Rolfe : New portable instrument for long-term ambulatory monitoring of posture change using miniature electro-magnetic inclinometers, Med. & Biol. Eng. & Comput., 32, 357/360 (1994) 5) W. Murata, S. Tanaka: An analysis of activity in disabled elderly by posture measurement and the relationship between their consciousness of their occupational roles in family (in Japanese), Annual Report of Gerontological Research, 14, 11/18 (1998) 6) S. Miyazaki: Long-term unrestrained measurement of stride length and walking velocity utilizing a piezoelectric gyroscope, IEEE Trans. Biomed. Eng., 44-8, 753/759 (1997)
-566-
