Research Of Moving Object Detection Algorithm In Video Surveillance System

With the development of computer network and modern economy, video surveillance has widely applied to every aspects of our life. It is an important measure to protect the public security, combat crime and prevent natural disasters. As the first step of video surveillance, moving object detection has great research value and meaningsAccording to the requirements of Guangxi Key Scientific and Technological Projects "High-Resolution Monitoring Control System of Communications Base Stations", this article has made comprehensive and profound study in background image model construction; moving object detection; shadow elimination; illumination change control and so on under the static scene. The following points are some achievements of the study:(1) Since the update rate of traditional Gaussian mixture background model cannot adapt to changing scenes, this article aims at solving this problem in the framework of traditional Gaussian mixture model by using matching probability of each gauss to calculate learning rate independently, not only updating the maximum Gaussian model distribution which matches with new pixels but also updating all eligible Gaussian model distribution at the same time, which ensures each Gaussian distribution of the system can be effective at all stages of learning. The simulation results show that the improved algorithm can effectively adapt to scene changes and improve the effectiveness and reliability of detection.(2) Proposed and implemented the solutions for real-time detecting the illumination change in monitoring scene. Illumination change has an important effect on detecting moving target. For instance, without processing illumination change, background model may mistake most of the scenes for moving target and prevent the subsequent step of system from normal functioning. This article puts forward a method of using video image histogram to measure the variation of illumination, then distinguishing the illumination change by measuring the absolute difference between illumination parameter of two successive video images on the basis of histogram entropy algorithm. The illumination parameter is introduced into the background model and the background model is self-adaptive updated according to the illumination parameter to eliminate the impact of illumination changes. This operation ensures the practicality of monitoring system.(3) The dynamic choice of Gaussian mixture background model distribution. The traditional Gaussian background model established Gaussian model in each pixel, which needs huge calculation. In order to improve the system's real-time detection, single or a few Gaussian background models can be set up for pixels in the static area and several adaptive Gaussian models for pixels in the dynamic area. (4) According to the contour information of moving targets, an effective, fast shadow elimination algorithm is proposed to eliminate the shadows through analyzing the improved Gaussian mixture model and the test results of gradient-background subtraction algorithm. The basic idea of this shadow elimination algorithm is to set the areas outside the closed target contour or its junction contour lower than the threshold value as shadow regions; while areas inside the closed target contour or its junction contour greater than the threshold value as foreground objects.