Implementation And Research Of Scale-Adaptive Object Tracking Method Based On Correlation Filters

Accurate scale estimation of a target is a challenging research problem in visual object tracking.Most state-of-the-art methods employ an exhaustive scale search to estimate the target size.The exhaustive search strategy is computationally expensive and struggles when encountered with large scale variations.In order to achieve a good balance between speed and accuracy,thesis investigates the problem of accurate and robust scale estimation based on correlation filters.Thesis propose two methods: divide-and-rule object tracking base on correlation filters and context constraint object tracking based on correlation filters.Specific work is as follows:1.In order to solve scale variation,this thesis paper proposes a menthod based on divide-and-rule strategy.First,training a filter,the scale information is added to the process of the target response,so that the candidate target not only has the position information,but also the scale information.Secondly,according to the scale information obtained in the previous step,use multi-resolutionthe strategy to improve the accuracy.The experimental results show that the method can achieve real-time and solve scale variation.2.The conventional correlation filter tracking algorithms,they do not have the ability to solve scale variation.Although these algorithms achieve good accuracy,it has a large time cost.This thesis paper proposes a simple and fast scale-adaptive object tracking base on correlation filters.Firstly,constructing a scale-adaptive search strategy to reduce computation.The strategy designs the scale selection factor to sample the current frame.Secondly,the proposed method investigates context information to a training filter.This operation improves the accuracy of the tracking algorithm.The experimental results show that the proposed algorithm can achieve good balance improves the tracking speed while maintaining good tracking accuracy and improving the overall performance of the algorith balance between speed and accuracy.Compared with other methods,the proposed method has higher accuracy and robustness when encountered with large scale variations.