Complex Activity Recognition System Based On Cascade Classifiers And Wearable Device Data

This paper proposes a system for recognizing human complex activities by using unobtrusive sensors such as smartphone, smartwatch and bluetooth beacons. The method encapsulates two classification stages. The former is composed of two parallel processes: the Main Activity Detection (MAD) and the Room Detection (RD). The latter implements the Complex Activity Detection (CAD) process by exploiting the outputs of the first stage and the accelerometer data of the smartwatch. The cascade classification approach that combines the room detection with the main/ complex activities recognition task constitutes the novelty of the work. Preliminary results demonstrate the reliability of the system in terms of accuracy and macro-F1 score.