Moving Object Detection Algorithm Based On Surveillance Video

In recent years,as the computer technology flourishes and becomes familiar with more ordinary people,smart devices gradually become essential to people's daily life.Therefore,the smart remote monitoring equipment emerges as time requires.By substituting computer vision for people's eyes,the technology liberate the labor force greatly;in the meanwhile,it makes the moving object detection algorithm even more important in the video.A good moving object detection algorithm can extract the moving object from the video accurately in real time,which is not only applied in the blocks where people live in and stations where they congregate,but also in fields such as public transportation,military and medical treatment and so on.Moreover,after abstracting the moving target from the video,we can also apply further operations,such as trace its path and analyze its behavior patterns.Therefore,abstracting the moving target swiftly but completely is the base of video analysis and video mining.In the following article,the benefits and drawbacks of some basic methods about basic image processing techniques and then the basic principles of optical flow,frame difference,and background subtraction,which are the traditional moving object detection algorithms,are analyzed.Then it introduces four methods of background modeling,which are mean value method,median method,kernel density estimation and SGM(Single Gaussian Model).This article aims at improving the drawbacks and generalize a more widely used and accurate moving object detection algorithm combing the advantages of each methods mentioned above through analyzing the whole process.This algorithm focuses on improving the inter-frame difference method and puts forward the improved five inter-frame differential,making the measuring of the moving target more accurate,which also solves the problems of the inconsistence of the margin.Moreover,it also improves the issue that the value in the GMM(GaussianMixture Model)can ' t be changed apart from the enormous calculation.Through calculate the video image separately,it reduces the calculation of matching algorithm greatly and raise a new scheme,which can self adaptively decide the value based on the frames that previously flowing by in the video.In the beginning of the video,it can complete the background modeling swiftly and then reduce the value gradually to prevent noises better.Because the frame differential method isn't sensitive to the alter of light,the results will be greatly discounted towards the rapidly and constantly altered dynamic background;while the GMM exactly applies to complexly changed dynamic background changes and extremely sensitive to the alter of light.Therefore,this algorithm absorbs the advantages of both methods,combining the improved five inter-frame differential and improved Gaussian Mixture Model,which completes the moving object detection algorithm jointly.