Research Of Moving Object Detection And Tracjking Based On Interest Points

Moving object detection and tracking in the image sequence is an important part in the research of computer vision. It is widely used in the field of aerospace, intelligent robot, automatic monitoring systems, medical image analysis and video compression, etc.Digital image is polluted by pulse noise during its acquisition and transmission (Digital process), the quality of image filtering will directly impact the effect of the following processing such as image segmentation, edge detection and feature extraction, etc. The salt-and-pepper noise is an important class of image noise, the traditional method is short at preserving image edge details. To solve this problem, an improved algorithm is proposed. The pixels of the image are classified into signals and possible noise based on amplitude distribution property of the salt-and-pepper noise. Then the possible noise pixels are judged if they are really noises according to the direction information and average variance. The edge pixels and the noise pixels are filtered by different methods. Besides, the idea of iterative filtering introduced can preserve image edge details as well as effectively remove image noise. The simulation results show that the accuracy and adaptability of image filtering can be improved greatly.In the moving object detection, Accumulative Difference needs accumulate many frames in order to detect small movement and slow changes, so it will detect a lot of background texture as well as has large overhead of time and space. For this, a new weighted Accumulative Difference method is proposed. First the least frames used to be accumulated are determined, and then sets the weights according to the impact of frame distances to Accumulative Difference picture. The weights setting should not only detect small movement and slow changes but also effectively suppress detecting noise pixels. The Weighted Accumulative Difference Method reduced the frames to be accumulated, increased the speed of moving object detection, as well as highlighted the differences between different frames of the accumulative effect. Experimental results show that The Weighted Accumulative Difference Method can effectively detect small movement and slow changes, and can greatly enhance accuracy and efficiency of moving object detection.In the moving object tracking, the Interest Points based method is employed. Every Interest Point is represented by a vector includes the characteristics of gray-scale and geometric characteristics. The efficiency of moving object tracking is improved by the Maximum Velocity Principle to determine the scope of interest points matching. When Interest Points are matched frame by frame, in order to avoid retrospect cased by mis-matching, this paper designed a non-retrospective mis-matching detection algorithm with the Consistency Principle. Real-time object tracking and robustness is greatly improved by solving fault matching and Many-to-one matching before the final match. When an object is determined to be occluded by the Occlusion Criterion, establish a tracking rectangle for each object, calculate the position of centroid using the points match. When an object is occluded completely, identify the controversial points in the overlap rectangles with a certain object, and then using the Kalman filter to predict the centroid. After the position of centroid is determined, the tracking rectangles can be regained according to the geometric relationship between the centroid and their tracking rectangle. Finally, this paper designed a processing flow to deal with object emerging, occlusion and disappearing in the process of non-normal tracking circumstances, which improved moving object tracking in the universality and robustness.