Monocular Body Pose Estimation by Color Histograms and Point Tracking

Accurate markerless motion capture systems rely on images that allow segmentation of the person in the foreground. While the accuracy of such approaches is comparable to marker based systems, the segmentation step makes strong restrictions to the capture environment, e.g. homogenous clothing or background, constant lighting etc. In our approach a template model is fitted to images by an Analysis-by-Synthesis method, which doesn't need explicit segmentation or homogenous clothing and gives reliable results even with non-static cluttered background.