Moving Object Detection Approach In The Condition Of Camera Motion

As one of the important contents of computer vision, moving object detectionapproach is the basis of object detection and tracking, involving a lot of knowledge,such as image processing, pattern recognition and artificial intelligence, and automaticcontrol. It has wide application prospect in security-monitoring, intelligent traffic,national defense and military, medical and so on. With the further development andwidely used of moving object detection technique, and people put forward morerequirements towards it, the tendency of moving object detection approach from thefixed camera condition to the camera motion condition is inevitable, in the research ofcomputer vision. This paper introduces a kind of moving object detection approach inthe condition of camera motion. Main research contents, results, and innovations are asfollows:(1) Motion estimation algorithm in the condition of camera motion is studied indetail. For video motion, the background image is constantly changed, which isseriously affected the accuracy of moving object detection. So as to solve this problem,combined with motion estimation and motion compensation technique, motionestimation algorithm based on non-motion region partitions is proposed in this paper.Corners are extracted by Harris in the reference image, and all in motion regionpartition are removed. Then, pixel points are matched with corners in the non-motionregion partition of the reference image, and motion vector is computed to compensatethe current image. Many groups of experimental results show that the algorithm canhave high level of precision and executive efficiency.(2) Motion object detection approach in the condition of camera motion is studiedin detail. On basis of motion estimation, and combined with connected regiontechnology, this paper proposes a kind of motion object detection approach based onedge information and connected region. This method uses Otsu’s method to get thethreshold and converts the difference image with binary processing. Then, K-nearestneighbor median filter method and morphological filter method are used to filterinterference noises in the image. Last, Moving object is detected by edge informationand connected region in the image. Many groups of experimental results show that theapproach can accurately detect moving object, and object has better integrity.(3) Using DirectShow, the video signal processing software is designed to get experimental results in many experiments of moving object detection approach in thecondition of camera motion.