Monocular Vision-based Human Motion Capture

Since human motion capture develops rapidly in virtual reality, multimedia, and man-computer Interaction, it has become a hot research topic in recent years. With the advancement of computer process capability and image manipulation technology, the human motion capture based on computer vision has been receiving more and more attention from researchers. Due to the low costs of obtaining human motion videos, human motion capture based on computer vision can reduce the costs of capture significantly and dig more potential information.This paper researches the human motion capture based on monocular vision. Firstly, Human motion is detected by background subtraction and post processing is applied. Secondly, the distinguishing trait of human body is tracked by using particle filtering. Then, the human body joints data is restored from two-dimensional space to three-dimensional space, and finally the human motion is reconstructed. It includes the following content concretely:1. To extract full human body with background subtraction, morphologic operations are used to remove the noise and fill the holes. A new shadow detecting algorithm is proposed to suppress the shadow of human body. The area of human body is located by human motion contours and handled by skeleton as auxiliary information.2. Human body joints tracking within the framework of particle filter is researched and the skeleton model which can used under two-dimensional space and three-dimensional space is constructed. The extraction and tracking of human body joints in video image sequence are both based on this skeleton model. When tracking the joints of human body, the joints of human body in first video frame are obtained manually and the joints of human body in other frames are obtained by particle filtering. Combining the low-level image information by thinning and the position of human body joints by tracking, the position of the rest of human body joints is restored.3. The 3D data is reconstructed by using the human skeleton model as prior information and the scaled orthographic projection model. Finally, the human motion are reconstructed. The results show that the method of this thesis can resolve the problem of motion capture based on monocular vision, and lay a theoretical foundation for future research and study.