3D Human Motion Reconstruction from Monocular Videos through Iterative Optimization

A novel approach on reconstruction of 3D motion of a human figure from a monocular video is proposed. By the method, a human stick model and the 2D feature points of each frame are obtained manually. The scale factor is computed and optimized in each frame. In the solution, an iterative optimization algorithm for a series of subsequence is also proposed to obtain the optimum solution. Finally, the 3D coordinates of each joint is computed to reconstruct the sequence of 3D human motion. Experiments on challenging videos including complex and highly dynamic motions show the effectiveness of the approach. This approach can be applied to real videos such as sports, movies and so on.