Research On Object Tracking In Multi-camera Based On Compressive Tracking Algorithm

With the development of Internet of Things technology,surveillance cameras have been widely covered in every corner of our daily life,video object tracking technology is also applied to different actual monitoring scene.A large number of target tracking algorithms are booming and applied to different scenarios according to their tracking performance characteristics.Recently,a research hotspot in signal processing,compressive sensing theory,is introduced into the target tracking domain,that is,real-time compressive tracking algorithm.This algorithm has good real-time performance and robustness because of its small calculation amount and robust tracking effect,and has been widely applied in reality.However,when target is severely obscured and the size of target changed intensely,the tracking process is easy to drift,even to lose.To solve those problems,a compressive tracking based on hypothesis testing and a multi-scale adaptive compressive tracking algorithms are proposed in this paper,they improved the tracking accuracy and robustness obviously.Because the single camera is limited by the region of monitoring and can not track the target effectively for a long time,applying the target tracking algorithm to the multi-camera will improve the flexibility,persistence and robustness of the target tracking,and make the target tracking technology get improvement effectively.In this paper,we combined the compressive sensing which are fast and robust and the non-overlapping multi-camera network object tracking system,and proposed a non-overlapping multi-camera network tracking system based on compression sensing.It includes five modules: single camera tracking module,multi-camera network scheduling module,target detection module,object re-identification in multi-camera module and target behavior trajectory analysis module.In the multi-camera scheduling module,a non-overlapping multi-camera network scheduling algorithm based on cooperative model is proposed.It based on the three parameters of cameras transition time,camera abutment relation,distance between object and camera,the scheduling model is effectively improved.In the moving object detection module,system used the three-frame difference method,a threeframe difference method based on MHI with higher accuracy is adopted.This method can extract the outline and complete region of the moving object and improve the accuracy of the detection Rate and effectively eliminate the lag and "empty" problems of the original three-frame difference method.System put the compressive sensing which are fast and robust into the field of object reidentification in multi-camera innovatively,effectively shortening the re-identification time.And an object re-identification algorithm based on compressive sensing and HOG cascade mechanism is proposed,based on the guaranteed accuracy,it improves the real-time performance and efficiency of object re-identification.The improvement and realization of each module are combined to realize the non-overlapping multi-camera network object tracking system based on compressive sensing.Experiments show that the proposed method can improve the accuracy and real-time performance of object tracking in complicated environment with large change of illumination and obvious change of size.It can achieve pretty good performance of real-time object tracking in cross-continuous view.