Research And Implementation Of Monocular Video-based Human Motion Capture

With the needs of community grow and the computer animation technology continues to develop, and the demand of motion capture material data grows, motion capture technology has been paid more and more attention. Traditional motion capture technologies need to use expensive equipment, and the sensors that attached to the body of the actors also limit their performance. These two shortcomings limit the development and application of animation technology. In recent years, the rising video-based motion capture technology solves the two shortcomings of the traditional technologies. This technology can track movement from low-cost video sequences and also let the actors' performance unrestricted. So it can reduce the cost of animation production and has very broad application prospects in intelligent monitoring, motion analysis and also other fields.2D tracking and 3D reconstruction are two main aspects of the monocular video-based motion capture technology. In 2D tracking, the human skeleton model is introduced, and the 2D tracking of human joints is achieved through the optical flow tracking method. In 3D reconstruction, the camera calibration is implemented with calibration template; the lengths of human skeletons are obtained as restrictive conditions through real-time manual labeling; the 3D resumption of the 2D information is achieved through the depth-first approach.For the self-occluding phenomena during the movement, the information of the occluded joint is resumed through the model and movement characteristics based approach. According to the prior human skeleton model, the information is resumed using the skeleton length and the continuity of movement as constraints. And then the unabridged motion capture data will be preserved as an user-defined format document. Then the data will be read into Maya and endued to the skeleton model to reconstruct the movement of the tracking target with MEL.Experimental results show that through this method the movement of the target in the video can be achieved accurately, and the 3D reconstruction of the 2D tracking information is also accurate, and the information of the occluded joint can also be resumed. So the unabridged motion capture data is obtained.