By employing tools in machine learning, computer vision, and image processing, the project will research the two important problems in visual object tracking and propose possible or better solutions. The two problems are sample-mislabeling and object appearance modeling. The project plans to solve the sample mislabeling problem in visual object tracking and to propose more effective and efficient method to model the object appearance based on the "relative space" concept and multi-subclass phenomenon of the object feature points. The project also plans to construct a software platform to test and evaluate tracking algorithms.
