Detection And Tracking Of Moving Objects Based On Compressive Sensing

Videos and images are very important information carriers for people. Videosurveillance system is convenient and visual, and is more and more widely used in allaspects of the society, like plaza, warehouse and so on. Detection and tracking movingobjects are the research highlights in computer vision. Detection and tracking movingobjects are the basic of sequent video understanding and analysis. As the time for “bigdata” is coming, people need more and more information, but the capacity of storagedevices and bandwidth for transporting data are limited. Since information is widelyspread, people are more concerned about privacy protecting. In addition, the situationof surveillance is various and resources are limited. So it’s difficult to propose arobust algorithm to solve above problems.Concerning above problems, a system for detection and tracking moving objectsbased on compressive sensing is proposed. This system firstly divides the frames ofthe video for fully using the relevance between the frames in video. Then use thebackground subtraction algorithm for the random sampled frames. Transport both thebackground subtracted frames and timed updated background frames to the decodingside, and this can make sure using less band width. Using Particle Filter to realize themoving objects tracking and Gradient Projection for Sparse Reconstruction algorithmto reconstruct the frames on the decoding side. It can be optional whether toreconstruct the original video to make full use of the resources. Moreover, theproperties of compressive sensing can add good security and robust to the data, andsimplify the coding side so that it can be used in a bad situation. Experiments provethat this system is compatible in many kinds of situations and better than originalsystems in speed, data size, privacy protecting and so on.Compressive sensing simplifies the steps of video compression and its maincomplexity is on the decoding side. It’s useful when resource is not sufficient or indistributed coding.