Research And Implementation On Moving Object Detection Algorithm In Video Images

The detection of moving objects in video images is an important research topic in the field of computer vision. Rapidly and accurately splitting out the moving target from the background is the foundation of the image for further analysis and processing. The technology of the moving target detection in video images has been used in many fields. Such as the automatic weapon targeting and wild environment monitoring. The moving targets in the complex environment can be quickly detected and located by the computer. Based on the Soldiers Confrontation System, this article did the research and optimization of the moving soldiers in the complex environment detection algorithm.The Soldiers Confrontation System is established indoor. There are multiple light sources, so that the ambient lighting is complicated. The Confrontation System is divided into two scenarios, and there are many shelters in both of them. Each of the scenario has an industrial camera and a screen set in the front of it. Through the camera, we can acquire the process of the soldiers’actions, such as moving or shooting. Through the screen, two groups of soldiers can see the movements of the other soldiers. Two groups of soldiers can shoot the other soldiers in the screen to do the confrontation. The outcome is decided according to the location of the shoot and size of the exposed area.Through the analysis of the moving target detection algorithm, we decided to use the Background Subtraction method to detect the moving soldiers in the Soldiers Confrontation System. The moving targets can be detected by subtracting the background frame from current frame. The optimization of the algorithm is proposed in this article. It is improved from two aspects, the efficiency and the accuracy affected by the interference.The algorithm can be improved from two aspects, qualitative and quantitative. After finding the moving targets. the size and the shape of them can be accurately analyzed in the images according to their positions. After that, most of the time is spent on the qualitative analysis, which reduces the time consumed by the quantitative analysis. An improved Nearest Neighbor Domain method is proposed in this article to do the down sampling of the video images, which can quickly reduce the scale of the images and improve the efficiency of the algorithm. This method can also maximize the disturbance information and improve the triggering sensitivity of the subsequent algorithm.In the real-time system, many factors leads to the noise in the scene, such as the camera shaking, light changing and shadow mask. During the detection of the moving targets, the abnormal caused by the environment changing can be easily mistaken as the moving targets. In this article, on the basis of Background Subtraction, an improved Surendra algorithm is proved to acquire the background information of the current frame. This method is used to update the background image in the Background Subtraction in real time, which can reduce the influence of the environment change. The accuracy of the algorithm can be improvedIt can be demonstrated in the experiment that the efficiency of the traditional moving target detection algorithm has been improved after the down-sampling and the background updating. Experiments show that using the down-sampling and background updating algorithm can improve the accuracy of target detection, which can also ensure the real-time performance of the algorithm.