Human Skeleton Reconstruction for Optical Motion Capture

In this paper, we propose a method of automatically estimating the pose of the underlying human skeleton to available in optical motion capture. These problems are addressed in the method: identifying markers over individual frames, predicting the missing marker locations and estimating the joint positions. Firstly, we identify markers according to local rigidity and space-time constraints of markers on the same limb segment in each frame. After that, based on small inter-frame motion, we predict the missing marker locations using the adjacent visible marker locations. Finally, we estimate the joint locations in two different scenarios, using a nonlinear optimization in the circumstance of less than three markers on one given limb segment. Results show that the approach is able to estimate the joint locations correctly, and our method is scalable in marker prediction and human pose estimation.