Research And Development On Background Subtraction Technology In The Complex Situation

The video surveillance is current research topic in computer vision, people pay more and more attention to this. The moving targets detection is the critical step in video surveillance, its aim is to extract moving regions from image sequences. The accurate segmentation is very important for post processing such as target classification, tracking and activity understanding, so the results and speed of moving targets detection directly affect the performance of video surveillance system. Regarding the detection of moving object in surveillance by the fixed camera, the most common and effective method is the background subtraction technology that can be used in machine vision surveillance applications such as traffic monitoring, bank surveillance. And in specific visual analysis and industry inspection study, background subtraction is also served as a good assistive technology. The background modeling and updating is the most important task in this process.Firstly this paper gives a comprehensive and detailed introduction on the process of background subtraction technology, and an analysis on the recent advances in background model algorithms. An adaptive learning rate running average background modeling method is proposed. An illumination detection algorithm based on the mean and variance and the moving objects detection algorithm by adaptive threshold technique are proposed. In eliminating shadow process, the static object and moving targets cast shadows are classified, and a rapid shadow detection method integrated RGB and HSV is proposed. Based on Visual C++ development platform, the image capture module with DG-400 image capture cards is realized to capture real-time images, and the functions of background updating and real-time moving object extraction based on mixed programming of MATLAB and Visual C++ are developed.The simulation and experiment results in the MATLAB environment show that the algorithm basically can achieve background updating real-time and accurately, and can identify the background shadow which should be added into background image. Based on Visual C++ platform, the developed image capture module can extract background images and moving targets.