Human Motion Tracking Based On Mul-objectives Optimization From Video

This thesis focuses on the automatic construction of three-dimensional human motion from video clutter which is one of the markers-less human tracking technologies. As we all known, human tracking is popular branch in the field of computer vision. The potential applications of this kind motion capture technique are, virtual reality, computer animation, video surveillance, medical diagnosis and so on. However, recovering3D poses and motions is always considered as an ill-conditioned problem since the depth is missing in the original2D image space or video scenes. How to track people effectively as well as accurately has still been a challenging topic in the research of computer vision.We proposed a3D human tracking framework based on multi-objection optimization algorithm. It considered human tracking as a multi-objection optimization problem and recovery of3D human pose as got best result of multi-objection in3D human pose space. While the construction of the aim objections is inspiring from generative an discriminative methods that is the two main stream in this field. We can got a few aim objections from the matching counter, gray, even3D pose getting from discriminative method. In final, the multi-objective optimization algorithm is using to get the best3D human pose.The main contributions of this thesis are follows:1. Multi-objective optimization is introduced in human tracking. This have avoided local optimization and reduced time consuming.2. We proposed a human tracking method based on multi-objective from multi-view. A few of aim objection constructed on multi-camera and reduced the collusion.3. Human motion tracking is considered as two functions co-optimization problem where the aim is to optimize the matching functions separately in3D an2D. Specifically the likelihood function is based on the discriminative pose prediction.This work is supported by the National Natural Science Foundation of China (No.61075041).