Object detection and classification is a fundamental and still unsolved problem. It has wide application in visual surveillance, robot navigation and complex video understanding. The key problem in this field is how to get the topological description of the object. Although significant progress has been achieved in this topic, it still cannot meet people’s ultimate expectation or goal. In this project, we will address this problem from three aspects: (1) building novel local structured feature descriptor. (2) Topologically modeling objects with local structured descriptors. (3) Reducing model’s complexity for accelerating training and testing. Finally, we will construct a system which can be easily applied for semantic video understanding.
(Sources of Funding: NSFC for the Excellent Young Scholars)