Research On Robot Object Tracking Method Based On Correlation Filter

With the rapid development of high-tech such as artificial intelligence,robots have new opportunities for progress and play an increasingly important role in people's daily life.Robot object tracking has high practical value,and it will become a basic function of the robot.In robot object tracking research,we first need to locate the target and then control the robot to follow the target.In the target positioning process,there are some methods such as depth camera,laser radar,binocular camera and other methods.The above methods requires expensive hardware costs,and failed to obtain satisfactory results.In the past few years,there have been sundry excellent tracking methods in the field of visual target tracking.These methods solve the difficulty of object tracking in different ways,and have achieved certain results.However,due to the particularity of the robot tracking problem,the current tracking methods cannot fully meet the target tracking requirements of the mobile robot.Based on the latest research in the field of vision object tracking,this paper studies the visual tracking algorithm suitable for a mobile robot platform.The tracking algorithm is transplanted to the robot platform to achieve stable long-term tracking.The method of object tracking for mobile robot must satisfy three requirements:real time,target scale and recoverable.Based on the latest research of correlation filtering theory tracking algorithm,this paper proposes the CFT(correlation filter tracking)algorithm.The algorithm has superior accuracy,can real-time tracking,and can obtain the target scale.In order to ensure that the algorithm restorable,this paper integrates the algorithm in the TLD(tracking-learning-detection)framework and proposes the TLDCFT(tracking-learning-detection correlation filter tracking)algorithm.The algorithm gets the advantages of both TLD and CFT two algorithms,and can meet the three requirements of robot target tracking.In order to objectively assess the effect of the above two algorithms,this paper experiments on the open dataset OTB2013.Experiments demonstrate that the proposed CFT and TLD-CFT algorithms can get better tracking results than Staple,DSST(discriminative scale space tracker)and TLD algorithms.The TLD-CFT algorithm has more significant improvement compared with CFT algorithm.In order to correlate the tracking algorithm results with the robot motion,this paper designs a robot motion command generation scheme.The TLD-CFT tracking algorithm is be transplanted into the robot Turtle Bot,to achieve a stable robot tracking system.The robot object tracking system is tested in a real environment.The results demonstrate that the robot object tracking system in this paper can stably track the target for a long time.