Application of nonparametric weighted feature extraction for an inertial-signal-based human activity recognition system

Abstract

This paper develops an inertial-sensing-based wearable human activity classification system and its associated human activity classification algorithm for accurately recognizing human daily activity. The proposed system used two inertial sensing modules, which are worn on subjects' wrist and ankle, to collect motion signals of human activities, and utilized the nonparametric weighted feature extraction (NWFE) algorithm to reduce the feature dimensions and improve the classification rate simultaneously. First, we integrated a microcontroller, a triaxial accelerometer, a triaxial gyroscope, and a RF wireless transmission module into the inertial sensing module. Next, the two inertial sensing modules are worn on subjects' wrist and ankle to form a wearable human activity classification system, and collect motion signals generated by daily activities. Subsequently, we developed an inertial-sensing-based human activity classification algorithm composed of signal preprocessing, human motion detection, windowing, feature extraction, feature reduction, and activity classification to recognize the types of human activities. In addition, we also utilized the NWFE algorithm to reduce the feature dimensions and computational complexity, and improve the classification rate. Finally, the experimental results show that the wearable human activity classification system and its human activity classification algorithm can recognize 10 daily activities with the classification rate of 90.5%.