Learning multi-level features for sensor-based human action recognition

This paper proposes a multi-level feature learning framework for human action recognition using body-worn inertial sensors. The framework consists of three phases, respectively designed to analyze signal-based (low-level), components (mid-level) and semantic (high-level) information. Low-level features, extracted from raw signals, capture the time and frequency domain property while mid-level representations, obtained through the dictionary learning method, learn the composition of the action. The Max-margin Latent Pattern Learning (MLPL) method is proposed and implemented on the concatenation of lowand mid-level features to learn high-level semantic descriptions of latent action patterns as the output of our framework. Various experiments on Opp, Skoda and WISDM datasets show that the semantic feature learned by this framework possesses higher representation ability than lowand mid-level features. Compared with existing methods, the proposed method achieves state-of-the-art performances.