The Research And Development Of Moving Object Detection Algorithm Based On Region Characteristics

With the development of society and the raising of people’s living level, researches and applications of intelligent video surveillance have received lots of attention. In intelligent video surveillance, the he main purpose is to detect, identify, classify and track the moving.objects, among which the moving object detection is the basis technology. Its results will subsequently affect the performances of the moving object identification, classification and tracking. Therefore, moving target detection as an important step in video surveillance, has a very important research value and significance.Existing moving object detection algorithms are mainly divided into two types:one is the moving object detection algorithm based on pixel gray values, which are characterized by only a single pixel grayscale without considering association between pixels adjacent impact resistance. Consequently, this kind of algorithm is susceptible to the illumination changes, noise and shadows and the detection result of it is inaccurate in general; the other type of algorithm is the moving object detection algorithm based on region, which divides the background images and detected images into blocks of equal size and extracts the features. But because of large error in segmentation, it is difficult to get the accurate shapes of the moving object. To cover the shortages of the existing moving object detection algorithms, this paper studies moving object detection based on pixel-level regional characteristics, in which the relevance of adjacent pixels is taken into account and the characteristic of an pixel is modeled by considering its surrounding pixels’characteristics. Our contributions are listed as follows.(1) A pixel-level moving object detection algorithm based on improved hu invariant moments was proposed. First, to address the problem of intensive computation in hu invariant moments, an optimizational algorithm is proposed and the feasibility of it is analyzed. Thus the amount of computation in each moment of hu invariant moments is reduced. Secondly, this paper uses the algorithm of improved hu moment invariants to extract the regional characteristics in pixels’ surrounding area as the pixel’s characteristics, and model the background with a single Gaussian model to detect moving objects. By comparing with the conventional single Gaussian model which uses pixel gray value as a characteristic value and mixture Gaussian model for moving objects detection, the experiment results show that our proposed algorithm is more effective in moving objects detection.(2) A pixel-level moving object detection algorithm based on Zernike moments was proposed. First, the principles and applications of Zernike moments are introduced in detail. And then, the changes of the moment’s eigenvalues in different situations in which there are moving objects passing through are analyzed. Correspondingly, the feasibility of applying Zernike moments into moving object detection is confirmed. After that, the single Gaussian area moving object detection algorithm based on Zernike moment is studied and the experimental results are discussed. Finally, the pixel-level moving object detection algorithm characterized by the hu moment and the pixel-level moving object detection algorithm characterized by the Zernike moment are compared. Experiments show that the pixel-level moving object detection algorithm based on Zernike moments has a better performance on detection accurate.(3) Finally, the above theoretical results are used in engineering practice. A moving object detection system based on DirectxShow and OpenCV is developed. The codes of our proposed algorithm are embedded into the OpenCV source codes to achieve a new detection engine. By comparing with the traditional mixture Gaussian model algorithm and K-neighboring algorithm existed in OpenCV, the performance of our proposed algorithm is verified.