Moving Target Detection And Recognition Algorithms

Moving target detection and recognition is the key point of the Image Process and Image understanding. With its application enlarged, its utility gets more and more important.The research interest of moving target detection and recognition is image sequence, i.e. video sequence. Detection algorithm aims to find out whether there are interesting target with processing the image sequence. And then algorithm will segment the target from the image. The purpose of recognition algorithm is to distinguish the target from swatch database, and also get the object characters.Based on the review and analysis of current methods on this domain, I present the whole framework of this system. And the same time, I have inaugurated some novel algorithm on the research of moving target detection and recognition. The major my research work and contributions are summarized as follows:1. Because the current algorithms do not pay more attention to the movement information with in inter frame. And this paper presents a novel black movement estimation with self-adapting 6 points distance based on inter frame. This algorithm well considers the movement information correlation within inter frame, and uses the diamond search in order to increase the correctness of the search rule. Because this algorithm uses the movement information of before frame, it can be implemented by parallel algorithm to improve efficiency.Experimental results show the algorithm constringed fast.2. How to solve the whole image variety due to the camera movement is a key problem and difficulty in the moving target detection. With research the current main algorithms of global estimation, this paper adopt the 6 points model of camera to compute this. The integer method only calculate the part black for decrease the time. Then the iteration algorithm can eliminate the infection from these moving target blacks. So we can get the parameters of the 6 points model of camera. Using this result every pixel can be complemented by bilinear interpolation. The simulation results prove this algorithm can deal well with the global movement cased by camera.