Model-based Human Motion Capture From Depth Map

The motion capture system can capture and save the human body's movements and appearance. As a part of intangible cultural heritage protection the preservation of human movement behavior can be achieved using motion capture system. While the traditional motion capture systems are inconvenient, and the device is usually expensive, in recent years, with the development of the TOF camera and other related hardware device, the depth image based motion capture technology has a growing number of applications for its convenience and low-cost.The purpose of this research is to explore the TOF camera-based marker-free human motion capture system. Based on the analysis of the related research and technology, this research first implements the current model-based marker-free motion capture method from depth map:The human motion and shape are modeled and the TOF camera is used to capture the image sequence to get the silhouette depth map of the actor, and then an iteratively nonlinear optimization method is used to transform the model to make it approach to the actor's real shape and finally the actor'motion is captured by the model. The main innovations of this dissertation are the error source and efficiency bottleneck of the original algorithm are analyzed, and further a variety of algorithms to improve the accuracy and to optimize the performance are proposed. A hierarchical and human parts-based pose estimation method is proposed to improve the accuracy and an octree-based method is proposed to optimize the human model to achieve a higher performance.