The Research On Detection And Retrieval For Moving Objects Based On Video

With the development of high technology, intelligent surveillance systems are used more widely. As the core technology in the intelligence surveillance, moving objects analysis based on video consists of the moving objects detection and retrieval, the object classification, incident detection, action identification and analysis, and so on, the detection and retrieval of moving objects is the foundation and key of it. In the detection and retrieval of the moving objects, the presence of shadows in an image can lead to misclassify the objects or merge the different objects, and bring the wrong results for the following advanced process, so it can't track the object accurately and give the correct understanding and description of the objects.Based on the summary and analysis of the relevant research works home and abroad, we make research on how to detect and retrieve the foreground and how to eliminate the shadow region in the moving objects detection and retrieval. The main research contexts and results are as follows:1. The background subtraction method is used to retrieve the foreground, and the mixture Gaussian model is used to model the background, during the modeling, the improved K-means algorithms is applied to improve the speed of background modeling;2. In the process of the background updating, the updating algorithms based on statistical average is used to update the background, compared to the traditional background updating methods, it improves the speed of the background updating;3. To solve the problem that the detection method based on Gaussian shadow model detects the shadow inaccurately under some situations, a novel algorithm based on the body color vector matching is presented to detect and eliminate the shadow. First, after the foreground is retrieved, the luminance test is used to reduce part the object region that has higher pixel's intensity in the foreground than the corresponding background. Secondly, each region's orientation line is calculated to prejudge the regions that contain the shadow regions, and the regions that contain shadow region are marked. Thirdly, the marked region's body color vector is calculated, and it is matched with that in the shadow database, after that, the shadow region can be obtained accurately.The experimental results show that the algorithm is robust under most situations,...