Research On Object Detection And Tracking In Dynamic Image Sequence

With the development of digital video technology, computer vision research techniques become more sophisticated, and as the important part of computer vision, moving object detection and tracking attracts more interest of researchers.This article focuses on moving object detection and tracking in dynamic background. First, we introduce the global movement and the local movement in image sequences, and then we can see the importance of the global motion estimation in image processing of dynamic background, then we introduce the principle and techniques of the global motion estimation. Additional, we analyze techniques of moving object detection in image sequences, and show the advantages and disadvantages of each method.Considering the defects of present methods and the feature of dynamic image sequences, we propose a novel method for moving object detection in dynamic image sequences based on global motion estimation, through which the dynamic background estimation can be easily done, and the affects of moving background are eliminated, and the experiment results show that this method works effectively and efficiently.Finally, we introduce the principle of moving object tracking, and classify present algorithms in a proper way, and then we introduce the principle of Kalman filter and its extended version. Having analyzed and compared the algorithm of Mean Shift and algorithm of Camshift, we find that Camshift is mainly a color-based algorithm, consuming less time than others, but when it forms a consistent area because that the color of the object is close to the color of background, it can not track the object well. So we propose the Camshift method of object tracking based on extended Kalman filter, it first estimates the position of moving object, and Camshift algorithm searches it on the basis of estimation results, more over, the computation caused by the searching can be greatly reduced, and the tracking is accurate. Given the experiment results, we proved the feasibility and effectiveness of the algorithm.