A Study on Human Activity Recognition Based on Acceleration-Angular Velocity Data and Link Mechanism
Yohei Sawai¹, Toshinari Kamakura²
¹Graduate School of Science and Enginieering, Chuo University, Bunkyo, Tokyo, Japan; ²Faculty of Science and Enginieering, Chuo University, Bunkyo, Tokyo, Japan

Recently human activity recognition and behavior analysis are often studied aiming at the service deployments to fields of the medical treatment and security. For observing human activities, previous works are basically grouped into two types of systems; one is based on video and wearable sensors and the other is on the motion capture system designed for analyzing human joint construction. The video and motion capture system is easy to consider joint construction, but it requires the laboratory room for filming 3D data. In contrast, acceleration sensor is easily handled in sense that we do not need any other facilities. However, it can not measure joint constructures and body sizes because the data obtaining from the sensor are only accelerations.

Hashimoto et al. (2008) proposed a method of estimation based on measurements of range of upper limb motion by using both a 3-axis wireless accelerometer and 3-axis wireless gyroscope. Kai et al. (2008) proposed a set of heuristics that significantly increase the robustness of motion sensor-based activity recognition with respect to sensor displacement. Ohmura et al. (2006) proposed a wearable sensing device called “B-Pack”, which is designed to capture nursing activities in real hospital. Shimada et al. (2006) proposed a survey report for the model fitting method to estimate 3D posture of articulated objects such as human body and hand.

In this article we propose a statistical method of estimating displacement angle of shoulder joint based on only one acceleration-gyro sensor. We model the displacements of the angle of the shoulder joint in each movement direction of the arm and estimate the parameters of the arm by nonlinear optimization techniques. We also show that our method can give a better classification in case of applying to gym exercise.

[1] Wataru Hashimoto, Hiroyuki Tamura, Haruo Noma, Masahiro Tada and Kiyoshi Kogure.”A Measurement Method for the Range of Upper Limb Motion Using a Wireless Accelerometer”. Wearable/Ubiquitous VR Conference 2008.

[2] KaiKunze, PaulLukowicz. ”Dealing With Sensor Displacement In Motion-Based Onbody Activity Recognition Systems”.UBIComp 2008 21-24

[3] Ren Ohmura, Futoshi Naya, Haruo Noma, Kiyoshi Kogure. ”A Bluetooth-based Wearable Sensing Device for Nursing Activity Recognition”.Wireless Pervasive Computing, 2006 1st International Symposium, IEEE 2006

[4] Nobutaka Shimada,Daisaku Arita and Toru Tamaki. ”Model Fitting of Articulated Objects”. Information Processing Society of Japan Vol.154 375-392 (2006)

Keywords: motion capture system; acceleration-gyro sensor; link mechanism; nonlinear optimization technique

Biography: Yohei Sawai currently works as a Graduate Student with the Satatistical Data Analysis Laboratory, Graduate School of Science and Engineering, Chuo University, In Tokyo. His area of interest is the Pattern Recognition of Human Activity.