Research Of Moving Object Detection Algorithm Based On Image Sequence

Moving object detection is a important researching content in computer vision,it extracts interested objects from the image sequence by employing methods of image processing,and has a powerful position in the field of intelligent video surveillance system.There are three main types of methods: Flow method;Frame difference method and Background modeling method.There moving object detection based on backgroud modeling mainly through two steps:Build the background model,partition image pixels and update the background model,and the modeling ways is various.It is updated real time in order to adapt to scence changes.It makes background modeling become a important research direction in the computer vision community.How to eliminate the complex background,dynamic noise,light changes and o ther disturbances in the real environment,and realize the robust and fast detection of the moving target has always been the problem that the researchers are trying to solve.This paper mainly studies the moving target detection technology in video s equence.(1)Three kinds of classical target detection methods have been studied:flow m ethod,background modeling method and frame difference method.Tow kinds of ba ckground modeling methods are introduced for background subtraction method.(2)The principle of CP3(Co-existence Probability based Pixel Pairs)algorithm is described in detail,and the advantages and disadvantages of CP3 algorithm are analyzed.The results of video detection in different scenes show that CP3 algorith m effectively overcomes the phenomenon of light mutation and background shaking,and has higher detection accuracy and robustness compared with Gaussian hybrid modeling GMM(Gaussian Mixture Model)algorithm.(3)In the view of the large computational complexity of CP3 algorithm,poor real-time performance and incomplete detection of moving target occlusion,a moving target detection method(CPS)is proposed,it’s named a integrate SLIC(Simple Linear Iterative Cluster)super-pixel and CP3 algorithm.Pixel Pairs).Firstly,the SLIC(Simple Linear Iterative Clustering)algorithm is used to segment the current frame into a super pixel set,and the singular Gaussian modeling of the linear correlation superpixel pair is carried out to construct a linear correlation background model.The influence of the number of superpixels on the target detection effect is studied,and the optimal value ofthe number of fast points is determined.The influence of the number of closely spaced pixels on the target detection results is studied,and the optimal value of the parameters is determined.(4)The improved algorithm is compared with GMM algorithm and CP3 algorithm in qualitative and quantitative comparison.The experimental results show that the proposed algorithm can help to quickly find and capture a more complete target of interest,and the algorithm can effectively deal with the problem of intermittent slow motion of the target,and the detection of the moving object is more complete.The improved algorithm is greatly reduced The modeling time,the run speed increased by 3times,the shelter is also a good robustness,so that the target detection comprehensive index Precision increased by 2.7%,the target detection comprehensive index Recall increased by7.4%,the target detection comprehensive index F-measure increased by 5.2%.