Study Of The Moving Target Recognition Based On Visual Attention

The research on computer vision has made a great progress recent years. Although it performs well in the structured environment, it does not perform well in the natural scene. On the contrary, humans are able to recognize interesting regions or objects in the cluttered environment easily.Visual attention mechanism solves the problem of computation bottlenecks. By limiting the amount of information need to be processed, it is possible to speed up search, recognition, and tracking. This paper focuses on the application of the recognition of the fast target in video sequences based on visual attention. The major works are listed as follows:(1) A framework of fast identification of the moving target is proposed which use Itti model to"scan"and use mean-shift algorithm to"attend".(2) Method based on saliency-map is suggested to modify the Itti's model which is the most popular to adapt the property and need of mean-shift algorithm.(3) In view of severe change of the targets'location between the frames, which made the method based on increment update difficult to tracking the targets, Itti's model is employed to supply the value of initial coordinates and kernel bandwidths which are very important to mean-shift algorithm's performance.(4) This paper also proposes a three-stage method employed in the situation that the target is not salient enough. Kernel density approximation is used to model the template.The results of the experiments validate the feasibility of the method.