Principal direction analysis-based real-time 3D human pose reconstruction from a single depth image

Abstract

Human pose estimation in real-time is a challenging problem in computer vision. In this paper, we present a novel approach to recover a 3D human pose in real-time from a single depth human silhouette using Principal Direction Analysis (PDA) on each recognized body part. In our work, the human body parts are first recognized from a depth human body silhouette via the trained Random Forests (RFs). On each recognized body part which is presented as a set of 3D points cloud, PDA is applied to estimate the principal direction of the body part. Finally, a 3D human pose gets recovered by mapping the principal directional vector to each body part of a 3D human body model which is created with a set of super-quadrics linked by the kinematic chains. In our experiments, we have performed quantitative and qualitative evaluations of the proposed 3D human pose reconstruction methodology. Our evaluation results show that the proposed approach performs reliably on a sequence of unconstrained poses and achieves an average reconstruction error of 7.46 degree in a few key joint angles. Our 3D pose recovery methodology should be applicable to many areas such as human computer interactions and human activity recognition.

A Review: Point Cloud-Based 3D Human Joints Estimation

“…The internal joints, such as the elbow, were fitted with an ellipse model to obtain. In the 3D point cloud, the Principal Direction Analysis (PDA) was used to estimate the main direction of the body part, and then the main direction was mapped to each part of the 3D model to estimate the human pose [ 83 ]. In the prescribed action set, a pose estimation using multiple random forests was proposed to enhance the results of motion analysis [ 84 ].…”