Research And Implement Of Moving Object Tracking With Correlation Filter

Object tracking technology has a wide range of applications in the fields of visual surveillance and autonomous driving,and it has received extensive attention from many scholars.Although the object tracking technology has made some progress in recent years,due to the increasing performance requirements in practical applications,as well as the interference of scale changes and occlusion.It has become a key point of research that how to improve the success rate,accuracy and anti-interference ability of target tracking.The problem of tracking failure is easy to be caused by the factors such as scale variation and occlusion.for the Kernelized Correlation Filters(KCF)in this thesis.The related algorithms of the correlation filters are deeply studied.And the improved algorithm and corresponding target tracking system has been designed.1.In order to solve the problem that the tracking algorithm of the kernelized correlation filter is prone to object drift in the face of scale changes,the method of separating window and scale pyramid is used to solve it in video object tracking in this thesis.A feature combination method for extracting global features of video images has been presented.And the fast scale adaptive correlation filter(FSACF)tracking algorithm has been designed.The FSACF algorithm guarantees the performance of the tracker while solving the problem of tracking failure caused by scale changes.2.In order to solve the problem that the KCF tracking algorithm is prone to model pollution when it encounters object occlusion during the tracking process,the tracking failure is caused.In this thesis,the color histogram is used to estimate the likelihood of the object,so that the hanning window with weakened boundary effect is optimized,and it is weighted and fused with the friendly gaussian window.In this way,the adaptive double window based on the color histogram solves the problem of model contamination,thus ensuring the tracking progress.3.In order to solve the problem that the KCF tracking algorithm is prone to object drift when facing the appearance change.The multilevel template strategy has been adopted to reasonably filter the object appearance information based on the average peak energy value,and retain the object appearance template information that is more useful for subsequent video sequences.The multilevel template group is used to classify candidate object by multilevel template strategy based on average peak energy value.The multilevel template adaptive tracking algorithm has been designed to improve the accuracy of the algorithm.4.The improved KCF algorithm model object tracking model system has been designed and implemented in this thesis,which imperforms real-time video object tracking system for selected object.