Research On Marker-less Human Palm Motion Capture

Human motion capture has become a quite useful technology which hasbeen applied to sports, computer interactivity system, and even medicalrehabilitation for motion analysis. In film industry and computer game, markerbased motion capture product are often employed to obtain high quality effectsrapidly. This allows the computer-generated characters to have the sameperspective as the video images from the camera.The traditional motion tracking methods need some reflective markersattached to human body and normally making some special restrictions tobackground in a scene such as blue screen background with a set of surroundingcameras captures the same object simultaneously.This paper presents a novel approach for human palms motion trackingwithout markers, which is different from the conventional marker based methods.The joints are estimated by adding depth and silhouette information. Combiningthe information from the depth and the color frames, a new matte is created. Adepth mask is generated by background subtraction and it can be used to handlethe edges of the palms with the corresponding RGB image. The segmentationstep does not make strong restrictions to the capture environment, e.g.homogenous clothing or background, constant lighting. Cam-shift algorithmutilizes color distribution to find target candidate and the background noise couldbe removed by combining motion detection. Since the changes between differentframes are so small Kalman Filter is used to predict coordinates of each joint soas to minimize the searching cost. Curve fitting is employed to reconstruct themotion trajectories of the discrete joints.Finally, the method proposed in this paper has been realized by a softwareplatform, visual studio2010, and the system runs at30frames per second on PCwith high accuracy on palm tracking. While the principle and feasibility of the method have been demonstrated, there are still several open issues to beaddressed in future work, eg. some complicated scenarios might cause accuracyproblems for human silhouette extraction which including human body movingwhile very close to objects in the scene such as chair, wall.