Multiview Cauchy Estimator Feature Embedding for Depth and Inertial Sensor-Based Human Action Recognition

摘要

The ever-growing popularity of Kinect and inertial sensors has promptedintensive research efforts on human action recognition. Since human actions canbe characterized by multiple feature representations extracted from Kinect andinertial sensors, multiview features must be encoded into a unified spaceoptimal for human action recognition. In this paper, we propose a newunsupervised feature fusion method termed Multiview Cauchy Estimator FeatureEmbedding (MCEFE) for human action recognition. By minimizing empirical risk,MCEFE integrates the encoded complementary information in multiple views tofind the unified data representation and the projection matrices. To enhancerobustness to outliers, the Cauchy estimator is imposed on the reconstructionerror. Furthermore, ensemble manifold regularization is enforced on theprojection matrices to encode the correlations between different views andavoid overfitting. Experiments are conducted on the new Chinese Academy ofSciences - Yunnan University - Multimodal Human Action Database (CAS-YNU-MHAD)to demonstrate the effectiveness and robustness of MCEFE for human actionrecognition.