On-Chip Activity Recognition in a Smart Home

This paper proposes a novel activity recognition system that is integrated into a smart home environment. It is characterized by low costs, high energy efficiency and low intrusiveness to increase the acceptance of the users. Activity recognition is performed locally on a single small-sized wearable device incorporating a microprocessor and a tri-axial accelerometer. After investigating on different feature sets and classification algorithms, the final implementation only considers five time domain features and a C4.5 decision tree classifier resulting in an immediate response. This wearable device is successfully integrated into an intelligent environment by Bluetooth Low Energy wireless communication protocol and openHAB as platform-independent integration software. The integration of the system into the home environment allows reactions of the home depending on the activity which enriches the life quality of the residents. Additionally, the system covers fall detection that enables the home to provide a fallen person with urgent support.