Learning Disentangled Representation for Mixed- Reality Human Activity Recognition With a Single IMU Sensor
Songpengcheng Xia, Lei Chu, Ling Pei, Zixuan Zhang, Wenxian Yu, Robert C. Qiu
Abstract
Together with the rapid development of the sensors technology in recent years, sensor-based human activity recognition (HAR) has shown promising performance using well-known supervised deep learning methods. But it remains challenging in a realistic scenario, i.e., limit number of labeled samples and sensors. This paper proposes a novel deep learning method to achieve accurate and robust HAR with only a single IMU sensor. Our contributions are two-fold: 1) Based on the SMPL (Skinned Multi-Person Linear) model, we build a large synthetic HAR dataset containing multi-modal measurements: acceleration and angular velocity, which were generated according to the forward kinematics; 2) We propose a multiple-level domain adaptive learning model with information-theoretically stimulated constraints to simultaneously align the distributions of low- and high-level representations of virtual and real HAR data. The proposed mutual information constraints encourage the neural network to learn a disentangled representation for the multi-modal sensing data. Comprehensive experimental results on three publicly available datasets demonstrate that the proposed method compares favorably with competing ones and has robust performance with variable labeled samples.