Solving occlusion problem in 3D human motion reconstruction

Abstract

One of the important problems in human motion analysis is the 3D reconstruction of human motion, which utilizes the anatomic pointpsilas positions. These points can uniquely define the position and orientation of all anatomical segments. In the proposed algorithm, joint tracking is performed based on a n*n block of DCTs coefficient (descriptor matrix). Then we correct the false tracked joint such as occluded joint using an energy function which can estimate false joints using the length of limbs and other joint location. We divided the problem into two parts: 1) Torso correction; 2) Hand and Leg correction. It leads to correct location of all joints even when the occlusion occurs for a part of time. Finally, 3D reconstruction is performed using Taylor method. Algorithm is robust in human turning and temporal occluded limbs. Experimental results showed the promise of the algorithm.