Markerless Full-Body Human Motion Capture

This work proposes a real-time markerless strategy to track human full-body movements for a pseudo 3D motion reconstruction using a single standard camera, at a low computational cost. This method allows a continuous depth warping of the tracked body parts based on pose recognition, and obtains realistic reconstructions when the considered actions are already known. A strategy to recognize combined actions, based on proper database storage of human motion patterns and poses, is also described. Experimental results show that taking advantage of the reconstructed poses it can be easily adapted to the different anthropometries of users without changing the known databases. This strategy allows more complex interactions between the subject and computer in human-computer interaction applications, as motor actions can easily be recognized in a wider set of poses, than those offered by holistic procedures.