Algorithmic Research On 3d Human Rigid Skeleton Extraction Based On Video Sequence

Human motion capture is widely used in intelligent supervision and control, human-computer interaction, virtual reality, computer 3D animation production, video-based physical training and gait analysis in medicine area, etc. Consequently, 3D human motion capture based on video sequence is worthy researching. Traditional human motion capture system needs to put makers, such as LED, on the subject, and transfers the position and motion direction of makers to the computer by sensors. Then, the computer analyzes the data, and the motion information of human are obtained. This system has some drawbacks: (1) it is very expensive, (2) it is hard to deal with the markers, (3) the subject will be hided from view, which makes it hard to match the markers with the control points, (4) it is hard to transfer the positions of markers into motion data. While, markerless human motion capture system usually uses several cameras to shot the people in the scene, reconstruct the 3D model of human from the synchronized video sequence, and finally, recovery the human motion including the positions of the joints and the angles of rotation of the body. Since markerless human motion capture researches straightly on the moving subject in video sequence, it is much more valuable.3D human skeleton, especially a rigid skeleton with the information of joins'central positions, represents the human motion status on a certain moment. Hence, 3D skeleton is significant for motion capture. In this paper, we focus on the study of 3D skeleton extraction algorithm in the markerless human motion capture system.Nowadays, many skeleton extraction algorithms for a coarse model, which has some noise, could not achieve satisfactory skeleton, let alone the joints'central positions. Aiming at this, we propose a skeleton extraction algorithm. Firstly, we reconstruct the volume model from the multiple-view synchronized video sequence. Secondly, we compute the curve skeleton of the volume model based on the theory of repulsive force field. Thirdly, we propose a criterion of linking curve skeleton, and link the different skeleton limbs using back tracking method. At this same time, we obtain the distance and angle threshold values by using binary search algorithm. Finally, after getting a smooth curve skeleton, we determine the joints'central positions of human skeleton using a priori information of human model. Results show that, this algorithm can obtain a desirable result as well as strong robustness, reduce the 3D skeleton extraction algorithm's sensibility to the noise of model boundary, and realize skeleton extraction automatically.