Research On Object Tracking Strategy With Fused Prior Knowledge

With the development of video surveillance technology,intelligent driving,intelligent transportation,intelligent city and so on,visual object tracking,a subresearch domain in computer vison,has attracted intense attention in both academia and industry,causing a large number of scholars and engineers to study the related algorithms and to achieve industrial application.Visual object tracking technology has made great progress after decades of research.However,during the tracking process,when the object in the video is under complex conditions such as motion blur,deformation,occlusion,out-of-plane rotation and so on,the tracking object will have a tendency to move away from the object.At this time,the tracking finally causes failure in the subsequent frames.In order to solve the above problems,this thesis proposes a visual object tracking strategy that fuses the prior knowledge of the object.It is mainly divided into the following two parts:(1)Acquisition of object prior knowledge.In the process of visual object tracking,the tracking object is annotated in the first frame of the video sequence.The information of annotated object is fully mined as the prior knowledge.(2)A visual object tracking strategy that fuses prior knowledge of the object to classical algorithms.When the tracking object box starts to move away from the tracking object,the object prior knowledge is used to perform the object search,and finally the performance of the classical visual object tracking algorithm is improved.In this thesis,the proposed object tracking strategy which fuses prior knowledge of the object is applied to the CT algorithm which is based on compressed sensing theory,the earlier correlation filtering algorithm MOSSE and the recent algorithm BACF.On the mainstream visual object tracking datasets OTB2013 and OTB2015,a comparative experiment is set up to verify the effectiveness and robustness of the proposed object tracking strategy by qualitative and quantitative analysis of the experimental results.