Jogging and Walking Analysis Using Wearable Sensors

doi:10.4236/eng.2013.55B005

Paper Menu >>

Journal Menu >>

Engineering, 2013, 5, 20-24

doi:10.4236/eng.2013.55B005 Published Online May 2013 (http://www.scirp.org/journal/eng)

Jogging and Walking Analysis Using Wearable Sensors*

Ching Yee Yong, Rubita Sudirman, Ahmad Hazwan Ab Rahim,

Nasrul Humaimi Mahmood, Kim Mey Chew

Infocomm Research Alliance, Faculty of Electrical Engineering, Universiti Teknologi Malaysia, Johor, Malaysia

Email: [email protected]

Received 2013

ABSTRACT

Gait analysis is a process of learning the motion of human and animal by using wearable sensor approach and vision

approach. This analysis is mainly used in medical and sports field where the study of body parts is crucial. 3-space sen-

sor is a sensor consists of accelerometer, gyroscope sensor and compass sensor, built in one device. In this study,

3-space sensor is used to collect data of walking and jogging motion, of a test subject running on a treadmill. Angular

velocity of the test subject’s arm and the angle of subject’s leaping motion are the two main components under investi-

gation. Data are analyzed and processed with Principal of Component Analysis (PCA) technique. This method aims to

combine and reduce the number of variables of the raw data. The Quiver function is used in order to generate feature

vectors for both motions. Furthermore, the output of the process was used to create a system that can recognize human

motion on any given data. The system is highly able to differentiate both of the motions.

Keywords: Accelerometer; Gyroscope; Motion; Wearable Sensor; Leaping

1. Introduction

Gait analysis is a study about movement of animal, or in

other words, a study of animal locomotion, with the aim

to analyze human motions, which is widely used in

medical purposes. This implementation is crucial to il-

lustrate the different body dynamics which depend on the

test subjects’ conditions, as stated in [1].

This analysis can be done by using two methods. The

first method, which is classified as vision based approach,

using video camera to record the movement of the sub-

ject. The second method is using wearable sensor, by

attaching a tri-axial sensor to the upper limb of the test

subject to generate sets of motion data.

In this paper, the second method is implemented.

3-space sensor is attached on the test subject to acquire

the motion data. Graphs were generated to represent the

orientations, rotation and acceleration (force) of the mo-

tion which are walking and jogging. A system is devel-

oped to recognize and differentiate these motions.

2. Research Review

Gait analysis is a very important study of human loco-

motive that give a great advantage in sports and medical

purposes. Zhou et al. [2] used multiple wearable inertial

sensors on upper limb motion tracking to give appropri-

ate assessment to the stroke patients before they are dis-

charged from the hospital after their movements are

comparable to the normal movement of human. Other

than that, Liu et al. [3] used wearable sensor approach in

gait recognition experiment, where the movement of a

subject was used to recognize that particular individual.

This is one of the application using wearable sensors as a

domain in biometric identification.

3-space sensor consists of 3 types of sensor, tri-axial

accelerometer, tri-axial gyroscope and tri-axial compass

which is built in one device [4] as shown in Figure 1.

Some experiment needed three types of sensor to evalu-

ate the entire motions component, but this device can

generate data for all components simultaneously in one

move by the subject.

Accelerometer sensor: this sensor can sense accelera-

tion by gravitational force and the acceleration of moving

body. Measured in g. 1g equals to 9.81 m/s2, which is the

value of acceleration of gravitational force. Gyroscope

*Motion classification using angular velocity and leaping angle. Figure 1. 3-space sensor axis orientations.

C. Y. YONG ET AL. 21

sensor: this sensor measure angular velocity in rad/s. Com-

pass: measures rotation angle with earth magnetic pole as

references. Hence, any magnetic or metal object might

disrupt the reading of this sensor, measured in degree.

3. Materials and Methods

3.1. Data Collection

A test subject was selected and his movement was evalu-

ated thoroughly to differentiate the motion of a subject.

The subject for this experiment is a healthy male indi-

vidual, with the age of 23, without medical record of

defect or other sicknesses. The test subject was attached

with the 3-space sensor on his upper arm, for jogging and

walking assessment. These motions were performed on a

controlled treadmill with wireless connection from the

sensor to the wireless dongle. Treadmill was used as a

medium for test subject to perform a regular motion with

lowest bias.

A preliminary speed testing was done before collecting

any data. The test subject needs to perform a normal

walk and jog on ground for a distance of 10 meter. Time

was stamped to record the period for the test subject to

cover up the activities. Hence, the test subject’s normal

walking and jogging speed was identified with walking

3.7 feet/second and jogging 6.5 feet/second. This infor-

mation was used as speed setting of the treadmill.

For walking motion, the test subject was asked to per-

form a normal walking on the treadmill with speed of 3.7

feet/second for a 10 meter, 15 meter, 20 meter, 25 meter

and 30 meter as a session. Then, every single session was

repeated 10 times for each of the different distance. Jog-

ging activity with speed 6.5 feet/second was performed

using the same characteristics of the walking activity.

Thus 50 sets of data for jogging and walking motion

were logged by the sensor were saved in .txt file.

3.2. Angular Velocity

The leg and arm are playing important role in human

motion. These two components are moving in a periodi-

cal wave while the test subject is performing jogging and

walking activities. The arm of the subject is the main

focus of the study.

While moving forward, the test subject is swinging his

arm forward and backward (rotation in y axis) and per-

forming rotation in vertical axis (z-axis), as shown in

Figure 2. Rotation in x-axis is very small for walking

and jogging and hence data for this are ignored.

The main thing of the analysis is to detect the positive

angular velocity for each swing of jogging and walking

activities and then calculate the mean value of every

swing. PCA is used to get the best view of the motion on

the graph after the variables were reduced as mentioned

in next subsection. Since the rotations of the arm during

activities are not perfectly aligned along the y and z-axis,

PCA was used to realign the graph to get the actual axis

of the moving arms rotations, as pictured in Figure 3.

3.3. Principle Component Analysis (PCA)

Principle of component analysis process shown in Figure

4 is a procedure in analyzing 2 or 3 dimensional set of

data, or to minimize the number of data by creating an

artificial point to represent the set of data. A randomly

100 points of data were selected from each session for

PCA analysis. The scatter plot graph shown 100 points of

distribution after PCA and each point was represented as

3 aspects (acceleration, angular motion and angle) for a

data collecting event, as written in [5]. Several steps were

performed such as covariance, covariance matrix, Eigen

vector and Eigen value to generate an artificial point for

a set of data for PCA variables reduction.

Figure 2. Rotation motion of test subject’s arm.

actual view

before PCA

after PCA

Figure 3. Rotation Perfect arm rotation axis (left), actual view & view from data (center), actual arm rotation axis rotated by

PCA (right).

C. Y. YONG ET AL.

Figure 4. Principle of Component Analysis.

3.4. Angle of Leaping Motion

Based on Tom F. Novacheck [6], in a complete cycle of

running or jogging motions, the runner will be airborne

twice, means that in running activity, there will be at any

instance where the test subject leap in order to get air-

borne. While in walking activity, there will be only

phases, where the test subject in a position called stance,

where the test subject weight is supported only by a foot,

and another one is when both feet touch the ground.

In short, leaping motion does not exist in walking mo-

tion and hence, this clear difference is used as the com-

ponent to differentiate both walking and jogging mo-

tions[7]. The leaping motion was measured using angle

of acceleration of the test subject.

Thus the second part of the analysis component is

leaping motion after the angular velocity. The leaping

motion was measured using quiver function in MATLAB

to analyze the angle of leaping and direction of leaping.

PCA was not used in this analysis in order to maintain

the original ground axis of the subject motion.

4. Result and Discussion

Figure 5 and Figure 6 show the best view of gyroscope

sensor after PCA variables reduction and direction re-

alignment for walking and jogging activities. The plot

shows that jogging data are scattered with a range larger

than walking activity. The direction and leaping angle for

jogging activity are recorded with a range higher than

walking activity. Result shows that jogging activity

brings bigger swinging angle and higher leaping dis-

tance.

Based on the value of mean of angular velocity and

leaping angle of test subject’s arm, a range is generated

for both walking and jogging activities as shown in Ta-

ble 1.

Figure 5. Scatter plot of acceleration of walking motion

(left), vector plot of acceleration of walking motion (right).

Figure 6. Scatter plot of acceleration of jogging motion (left),

vector plot of acceleration of jogging motion (right).

Table 1. range of angular velocity and leaping angle for

walking and jogging motion.

WALK1.1007 ≤ angular velocity ≤ 1.9119

Angular velocity

(rad/sec) JOG 2.5697 ≤ angular velocity ≤ 3.9663

WALK4.629 ≤ leap angle ≤ 16.10

Leaping angle

(°) JOG 38.0579 ≤ leap angle ≤ 62.3121

C. Y. YONG ET AL. 23

Through observation from the table, the angular veloc-

ity of test subject’s arm is faster in jogging compared to

walking. It means that the test subject swing his arm

faster in jogging compared in walking. Since the range of

angular velocity of walking and jogging are not overlap

with each other, this range is used to recognize and dif-

ferentiate both walking and jogging motion.

Same thing goes to leaping angle assessment. Jogging

activity has recorded a range higher than 35°.

The mean of angular velocity and the mean of leap an-

gle of the test data yield from the analysis are used to

evaluate the motion classification program represented

by flowchart in Figure 7. A system was developed based

on the range of angular velocity and leaping angle for

motion classification.

5. Conclusions

It can be concluded that a 3-space sensor is useful in

analyzing human motion since the device consists of

three different sensors which give advantage in measure-

ing the motion in several aspects simultaneously[8]. The

success of the data collecting session also put wearable

sensor approach one step forward of the vision based

approach in terms of the number of different measure-

ment can be done simultaneously.

There are two kinds of graph generated in analyzing

both motions: they are PCA scatter plot and quiver func-

tion graph. PCA scatter plot realigned the graph for the

best view in angular velocity and reduced nine variables

become three for better analysis. The quiver function in

MATLAB is used to generate the direction vector plot of

the given motion. The vector plot shows the increment

and decrement of the motion direction. This information

shows clearly every move for a swing of human arm.

Finally, the range of angular velocity generated in Ta-

ble 1 is used to develop a motion classification system.

The system was trained by using the recent collected data

and the process of increasing the amount of data is

planned to be done in the future in order to increase the

accuracy and specificity of the system.

Start

2.5697<=angular_velocity<=3.9663?

1.1007<=angular_velocity<=1.9119?

38.0579<=mean_elev<=62.3121?

4.629<=mean_elev<=16.10?

Display JOG

Display WALK

Display

UNIDENTIFIED

Finish

yes

Figure 7. Flowchart of motion classification program.

C. Y. YONG ET AL.

Figure 8. Motion classification system.

The classification system was shown as in Figure 8.

The GUI is developed to be user friendly as users can

easily load their file to the system within a click. The

system displays gyroscope best view plot, acceleration

gait plot (human arm swinging animation), and accelera-

tion vector plot (direction of swinging arm). The mean

value of angular velocity and mean value of leaping an-

gle of the motion are displayed at the bottom of the plots.

The classification result is shown at the bottom right of

GUI. The system will display “WALK” for walking ac-

tivity, “JOG” for jogging activity, and “UNIDENTI-

FIED” for unidentified motion.

6. Acknowledgements

A study of this magnitude depends on the hard work and

commitment of many professionals, and we are pleased

to acknowledge their contributions. The authors are

deeply indebted and would like to express our gratitude

to the Universiti Teknologi Malaysia for supporting and

funding this study under Research University Grant

(QJ13000.2636.05J69) and MyPhD Scholarship Scheme

from Ministry of Higher Education (MOHE). Apprecia-

tion also goes to Muhammad N. Hafizuddin Zainudin in

assisting the data acquisition.

REFERENCES

[1] L. Tao, I. Yoshio, S. Kyoko and M. Haruhiko, “Devel-

opment of Wearable Sensor Combinations for Human

Lower Extremity Motion,” Proceedings 2006 IEEE In-

ternational Conference on analysis in Robotics and

Automation, ICRA 2006.

[2] H. Zhou, T. Stone, H. S. Hu and N. Harris, “Use of Mul-

tiple Wearable Inertial Sensors in Upper Limb Motion

Tracking,” Medical Engineering & amp; Physics. Vol. 30,

No. 1, 2008, pp. 123-133.

[3] R. Liu, J. Z. Zhou, M. Liu and X. F. Hou, “A Wearable

Acceleration Sensor System for Gait Recognition,” Pro-

ceedings of the 2nd IEEE Conference in Industrial Elec-

tronics and Applications, 2007.

[4] 3-Space Sensor, User Manual, YEI Technology.

[5] J. Shlens, “A tutorial on Principal Components Analysis,

Version 2,” Salk institute for Biological Studies and In-

stitute for Nonlinear Science, University of California,

San Diego, 2005.

[6] T. F. Novacheck, “The Biomechanics of Running,” Gait

& amp; Posture, Vol. 7, No. 1, 1998, pp. 77-95.

doi:10.1016/S0966-6362(97)00038-6

[7] M. Hanlon and R. Anderson, “Real-time gait event detec-

tion using wearable sensors,” Gait & amp; Posture, Vol.

30, No. 4, 2009, pp. 523-527.

doi:10.1016/j.gaitpost.2009.07.128

[8] A. Minami, H. Teruo, O. Tomoyuki, K. Kazuyuki, W.

Kajiro and K. Yosuke, “A Study on Gait Analysis by

Measuring Axis Rotation Based on 3D Magnetic and

Acceleration Sensors,” Proceedings of SICE Annual Con-

ference 2010.