Accurate 3D Human Pose Recognition via Fusion of Depth and Motion Sensors

Abstract

 Abstract—In this work, we present an accurate 3D human pose recognition (HPR) work via multi-sensor fusion. Lately, 3D HPR is widely performed using a depth imaging sensor, but this approach has limitations: 1) orientations of body parts cannot be accurately recognized and 2) it suffers from occlusion. To achieve an accurate and stable recognition of human poses in real-time, in this study, we propose to use inertial measurement units (IMUs) which are used to estimate the orientation of body limbs and solve the occlusion problem. Via fusion of depth and IMU sensors, our results demonstrate significantly improved 3D human pose reconstruction: our results show the accurate recognition of twist and location of the arms even under occlusion. Our presented approach could be critical if 3D HPR is to be used for medical applications such as musculoskeletal analysis via in 3D as demonstrated in this study.