The Research Of Object Tracking Based On Correlation Filter

With the popularization of Artificial Intelligence and Big Data,Computer Visual has plays an indispensable role in daily life.Object Tracking is an important means from the underlying information processing to high-level content,analysis in computer vision.It has important application in the fields of human-computer interaction,robots,driverless vehicles unmanned driving and intelligent video surveillance and others.Object tracking is the analysis of a given target in the video frame sequences,and the target is predicted to appear in the image frame.Based on object tracking algorithms involving correlation filtering,tracking algorithms are designed under complex environment in this thesis,which give consideration to both accuracy and real-time performance.With tracking effect increased,they can still run fast enough to keep real-time requirements.The existing correlation filtering trackers,e.g.kernel correlation filter(KCF),have not made full use of background environment surrounding object.In this thesis,by restraining local background information,adaptively updating scale of target search area and introducing detector,under the basic requirements for real time,algorithms are designed to improve the robustness of object tracking.Main works of this thesis are as follows:(1)A tracking model(RICF)restraining local background information is proposed for improving the object tracking effect effectively under the situations of motion blur,dramatic appearance change in a complicated environment.Specifically,during in the stage of training,it protrudes the object area by regressing four same size patches in the left,top,right and bottom of the tracking box to special background values.(2)A further correlation filter model of adaptive search window(RIACF)is proposed.In RICF,when the tracking object similar to some background area,or exactly in the search window border,drift phenomenon usually happens because the algorithm usually misjudges target characteristics as background.Therefore,an adaptive search window correlation filtering model(RIACF)is further proposed for improving the effect of object tracking by updating the object search area with scale threshold.(3)Based on the RIACF,an online detector in tracking model(DRIACF)is introduced to re-detect the current frame for finding the object again under the situation of lower similarity value,which is taken as the initial object for tracking in the following.It can be used for solving the challenges of object rotation and reappearing better.