Human Action Recognition using Acceleration and Physiological Data in Real-Time

C. Morita, M. Sato, and M. Doi (Japan)


Data Mining, Decision Tree, Wearable Sensor, Context Awareness, Ubiquitous Data.


This paper describes human action recognition using accel eration and physiological data. The data is obtained from a wristwatch-type sensor, called "LifeMinder" R . This sen sor measures 2-axis acceleration, the pulse rate, the gal vanic skin response (GSR), and the skin temperature. The purpose of the study is to enable recognition of general ac tions such as "walking" or "having a meal" in real-time using the sensor data. Recognition of such actions is useful for healthcare management services and medical treatment support. To enable recognition of actions, we extracted fea tures from the sensor data and constructed a decision tree using the features. We achieved recognition accuracies of over 96% for the actions by the appropriate feature extrac tion.

Important Links:

Go Back