Research On Correlation Filter Tracking Algorithm Based On Target Representation Enhancement

In recent years,with the continuous construction of smart cities,target tracking has become more and more important.Visual target tracking needs to locate and track the target in each frame of the video sequence.Complicated environment and target changes are the difficulties faced by target tracking technology.Therefore,the key task in the field of visual target tracking is to study target tracking algorithms with high accuracy and stability.Correlation filtering tracking algorithms have received extensive attention and research from scholars because of their high speed and high accuracy.However,the existing correlation filtering tracking algorithms still have the following problems:First,the boundary effect will make the training sample deviate from the true representation of the target,and the second is the lack of distinguishability of the features used for target description in complex scenes.These two problems will cause the trained filter to reduce the discriminative power of the target,and then affect the target tracking accuracy.Therefore,how to improve the accuracy and stability of correlation filtering tracking algorithms is a problem that needs to be solved urgently.In response to the above problems,this article starts with target representation enhancement,and improves the robustness and accuracy of the correlation filter tracking algorithm by studying boundary effect suppression and feature enhancement.The main work of this article is divided into the following two points.(1)This paper conducts an in-depth analysis of the cosine window that alleviates the boundary effect.It is not conducive to the relevant filter to learn complete and significant target information in the scale change scenario,and then proposes an adaptive space that can follow the target scale and shape changes.The regular window enables the regular window coefficient to adapt to the target size change,thereby better reducing the influence of boundary effects,enhancing the integrity and saliency of target sample features,and improving the discrimination performance of the filter.(2)This article conducts an in-depth analysis of the distribution characteristics of the response channel of the visualization filter.The response channel corresponding to part of the characteristic channel of the sample is beneficial to target positioning,while other channels will interfere with target positioning,and then the channel attention mechanism is proposed for sample characteristics.Screen all channels of,strengthen the contribution of the channel where the important feature is located,and suppress the influence of the channel where the interference feature is located on the filter training performance.In order to reflect the advantages of the algorithm in this article,this article selects the public data sets OTB100 and TC-128 to test and evaluate the algorithm.The test results show that the algorithm proposed in this paper effectively enhances the expression ability of target features under the scene with obvious scale changes or similar background interference,thereby improving the accuracy and success rate of tracking.In addition,this paper also highlights the superiority of this algorithm by comparing it with other correlation filtering algorithms that deal with boundary effects or features.