The Research On Moving Target Positioning Based On Video

Moving Target Positioning based on video is to extract moving targets from theontinuous frames based on video, and then to position the moving target preciously.As an important research technique of Computer Image Processing and MachineVision, it involves many areas such as Pattern Recognition、Artificial Intelligence、Automatic Control. Meanwhile it has a profound impact on the fields such asIntelligent Control, Robot Navigation, Military Guidance, Intelligent Transportationand Medical Image Analysis.Moving Objects Positioning mainly include two aspects: Moving TargetDetection and Moving Target Positioning. Moving Target Detection means to detectmoving targets from contious frames based on video. Moving Target Positioning is tolocate the position preciously from the Moving Target detected before.Although many researchers have studied on moving targets detecting andpositioning based on video,but there is no no specific algorithm or method can beused in any complicated scene so far.The complexity of the video scenes and theuncertainty of the status of the moving target and the shadow of the moving targetmay generate serious impact on the result of moving target detecting and positioning.So the research on moving target detection and positioning still has importantpractical and theoretical values.We propose a new algorithm to detect and poison the moving target which cansolve the problems above on the basis of using three-frame differencing, backgroundsubtraction, Gaussian Mixture Model and other classical methods. The algorithmmainly includes two aspects:1、Moving Target Detection, although we can get a good result when we useGaussian mixture model to detect moving targets, but there still exist some problems.On one hand, when the status of moving targets suddenly changed we may get wrongtargets and we may miss some targets. On the other hand, the effect of suppressioningshadows of moving targets is very poor when we use Gaussian mixture model. Inaddition, we need to model every pixel when we use Gaussian mixture model, thisrequires a large amount of calculation, which may produce a negative impact onreal-time requirement of video surveillance systems. In this paper, an algorithm formoving targets in video sequence detection based on Gaussian mixture model andthree-frame differencing is put forward to improve the deficiency of foregrounddetection based on Gaussian mixture model. First we can get the foreground and backgroung based on Gaussian mixture model. Then we can get another foregroundby using the current frame minus the background we got above. Finally we can obtainthe precious outline through three-frame differencing and edge detection, we can get aforeground by filling the outline. We can get the final result by doing and operationfor the foregrounds we got from the above three steps. In addition, we can use newupdate strategy to model the background faster, we adjust the updating speed of modelparameters according to the stability of each pixel in frames to reduce thecomputational complexity and improve the speed of the algorithm.2、In Moving Target Positioning, in this paper we get the position of the movingtargets when we can detect them accurately. Moving targets positioning is to label themoving tagets in the video accurately and get the coordinates of the central locationand the center of gravity.The algorithm can be simulated successfully by using Microsoft Visual Studiodevelopment platform based on Windows environment in this paper. From the actualeffect we got, the improved Moving Target Detection and Positioning Algorithm isfeasible and we can get accurate result.Compared with the conventional algorithms,this algorithm is better in eliminating interference factors of complex scenes andadjusting the sudden change of the state of the moving objects and eliminating theinfluence of shadows.