Human Motion Enhancement via Joint Filtering and Learning Approaches 

Abstract: Vision-based human motion capture is an important and active research topic in the field of computer vision and pattern recognition due to its wide applications. Due to recent development of low-cost RGB-D or depth sensors, depth-based motion capture (D-Mocap) has gained much research interest which has some advantages and benefits over traditional Mocap technologies. However, current D-Mocap suffers from low precision and poor reliability due to the self-occlusion problem and the limitation of depth sensing. This motivates many recent studies to improve the quality of D-Mocap data for possible precision-demanding applications. Specifically, we will study both filtering and learning approaches that have complementary nature for human motion enhancement. We present two frameworks that share two key elements in a different way, joint-level kinematics and skeleton-level anthropometrics. We first present a particle filtering framework that incorporates a new Tobit particle filter (TPF) to characterize the joint-level censored measurement and a differential evolution (DE) algorithm to enforce skeleton-level anthropometric constraint. Then we propose a filter-assisted autoencorder learning framework which integrates a general motion manifold learned from large set of high-quality Mocap data and the Tobit Kalman filter (TKF) to capture joint-level kinematics. The two methods are thoroughly evaluated on both simulated and real-world D-mocap data, and experimental results show that the quality of D-mocap data can be improved significantly by more than 50%, making D-Mocap a promising and affordable human motion analysis tool for various clinical applications.

Bio: Guoliang Fan received the B.S. degree in Automation Engineering from the Xi'an University of Technology, Xi'an, China, the M.S. degree in Computer Engineering from Xidian University, Xi'an, China, and the Ph.D. degree in Electrical Engineering from the University of Delaware, Newark, DE, USA, in 1993, 1996, and 2001, respectively. He is the Cal and Marilyn Vogt Professor in Engineering with the School of Electrical and Computer Engineering, Oklahoma State University, Stillwater, OK, USA. His research interests include Image Processing, Pattern Recognition, and Computer Vision. Dr. Fan is a senior member of IEEE and he served as an Associated Editor for IEEE Transactions on Image Processing from 2014-2018. Currently, he is an Associate Editor for the IEEE Journal of Biomedical and Health Informatics and EURASIP Journal on Image and Video Processing. He was the recipient of the Young Alumni Achievement Award from the Department of Electrical and Computer Engineering, University of Delaware in 2015.