Computer Vision Optimization Problems Using Machine Learning

Many Machine Vision problems may be framed as solving a difficult optimization problem. We intend exploring the use of new advanced optimization techniques in some specific popular tasks, particularly human motion recovery from video sequences, and understanding the structure of articulated and dynamic scenes. Our recent work on constrained optimization, along with research on techniques such as Second Order Cone Programming has shown the benefit of applying advanced optimization methods to common Vision tasks. In addition, Machine Learning techniques have been used in other fields to learn optimization parameters. We will introduce such techniques to Vision problems, with expected major improvements in capability.