ActDetector: Detecting Daily Activities Using Smartwatches

Detecting daily activities is helpful for health care and clinical medicine. In this paper, we present ActDetector, a smartwatch based application which detects 8 common daily activities, including sitting, walking, running, going upstairs, going downstairs, eating, driving and sitting in a vehicle. By leveraging the built-in sensors on smartwatch, a multi-level classification system is proposed which considers both detection accuracy and energy efficiency. ActDetector is designed to work unobtrusively, no matter on which wrist the smartwatch is worn. We have implemented ActDetector on Sony Smartwatch 3 and evaluated its performance in real experiments involving 12 users. Experimental results show that ActDetector is energy efficient and can detect the daily activities with high accuracy.