A Phone for Human Activity Recognition Using Triaxial Acceleration Sensor

Abstract

We propose an activity recognition system especially for the elderly using a wearable sensor module including a triaxial accelerometer. We have mainly tackled easy battery loss problem minimizing the efficiency decrease of the activity recognition. The proposed system consists of main modules; sensor module, gateway, and PDA phone. The sensor module is worn at the left side of waistband, and the embedded algorithm installed in a microcontroller of the sensor module manipulates the sensing data in order to reduce the transferring overhead which causes enormous battery loss. After that, the sensor module transfers them to the subject's PDA phone by means of the Zigbee. Then, the phone accumulates the data for a while and starts to classify them as an ADL (activities of daily living) like running, walking, standing, sitting, lying, falling, etc. As a classifier of ADL, the multi-layer perceptron algorithm is used, and we have achieved 95.5% of accuracy in inferring an activity out of 9 ADLs. It shows we can save the battery loss while maintaining the recognition accuracy similarly compared to related works.