Research On Moving Object Detection

In real life, moving images contain a large number of meaningful visual information. Moving object detection is the base of computer vision, motion, image analysis, human-computer interaction and intelligent video surveillance technology, which is widely used in the military, computer-aided design, aerospace, intelligent robotics and other fields, the test results have a direct impact to the follow-up treatment such as target location, tracking and behavior understanding. So moving object detection technology has a very important theoretical and practical application of significance. This thesis does research on the moving object detection in the background of a relatively stationary video surveillance system, mainly include:1. Describes some aspects of image pre-processing technologies, including: color space model, image denoising, image equalization, image edge detection, image morphological processing, and experimental validation of the technology.2. Briefly describes the image segmentation method, with fully descriptions for two kinds of moving object detection algorithm: frame difference and background difference, which are base of the threshold segmentation, including the principle of two object detection algorithm, besides the advantages and disadvantages between this two, and making the experimental verification of the algorithm. Considering the disadvantages of two object detection algorithms proposes some improved algorithms, which include motion detection algorithm based on Inter-frame difference method and edge information, and background update algorithm based on IIR filter. In addition, briefly introduces the knowledge of optical flow method, which is another commonly used algorithm for moving object detection.3. In order to solving the problems that easily lack to check when the object running slower or the size is smaller, proposes a method of motion detection combining time domain and spatial, base on existing method of moving object detection, which is the motion detection algorithm that combines of motion information and marker multi-measure watershed. This method based on information in the time domain, can get good results of moving object detection that combined with the watershed segmentation method. The result shows that the detection algorithm can detect object accurately and quickly.