Using compact distance energy model to recover human pose in markerless motion capture

Much of the research on markerless model-based human body motion capture always tries hard to construct accurate human body model and precise surface contour (e.g. using super-quadrics or ellipsoids to represent model limbs). Unfortunately, because of the main challenges of human body motion capture such as loose clothing, self-occlusion, image noise and background segmentation error, there are lots of error messages in the 3D reconstruct result of human body, which makes the efforts of these methods get very limited effects. In this paper, we propose a novel algorithm that uses a compact distance energy model (DEM) to recover the human pose which is robust to the inaccurate human body reconstruction caused by the factors mentioned above, and the DEM will be updated in EM framework in the process of motion capture to ensure that the DEM is compact to the body at any time. Experiments show the effectiveness of the proposed method.