Research And Improvement Of Target Tracking Algorithm Based On Correlation Filter

Target tracking technology is an important branch of computer vision.It is used in various fields such as TV monitoring,human-computer interaction,robot vision navigation,medical diagnosis,etc.Therefore,the development of target tracking technology promotes social production and improves the living standard of society.And the promotion of related academic research has very important significance.Although the target tracking technology has been widely used,the application scenarios that are faced have become more and more complex,such as lighting changes,background blur,scale changes,occlusion and other challenges.The existence of factors,therefore,is still a significant topic for how to improve the robustness and practicability of the target tracking algorithm.Since the tracking algorithm based on correlation filter can guarantee the robustness of the tracking algorithm while ensuring the target tracking speed,since the target tracking algorithm is proposed,many scholars have paid attention to the correlation filtering algorithms.The correlation filtering algorithm reduces the complexity of the operation by transforming the complex operation of the filter in the Fourier frequency domain,thereby realizing the rapid detection of the center position of the target,and re-sampling to update the filter online,ensuring the correlation of the filter target tracking.In this paper,the correlation filtering algorithms are deeply studied.In view of the difficulties in the current target tracking technology,some improvement methods are proposed in the previous research results.The following is the research content and innovation points of this paper:Firstly,the development status and research status of target tracking technology are summarized.Then the main framework of the target tracking algorithm in the target tracking technology and the main challenges are introduced.At the same time,the chapter layout of this article is briefly explained.Secondly,a target tracking algorithm based on context-correlation filtering of fusion features is proposed.When using the relevant filtering algorithm to achieve target tracking,the traditional correlation filtering algorithm often uses the cosine window to suppress the edge effect or expand the search range when the target is positioned(generally doubled).The background information that can be utilized in the target tracking algorithm is greatly reduced,so that tracking target drift often occurs during target tracking.Based on the shortcomings of the above traditional correlation filtering framework,a context-based target tracking algorithm is proposed,and proposed to use HOG features and CN features as input features to enhance the accuracy and robustness of the target tracking algorithm.It is proved by experiments that the improved correlation filtering is more robust than the traditional filtering algorithms.Thirdly,a target tracking algorithm based on convolutional neural network and correlation filtering is proposed.In recent years,deep learning convolutional neural networks have been widely used in imagery.The convolutional neural network model can extract the features of images comprehensively.However,convolutional neural network models often require a large amount of data and time for training,so this paper proposes to use an online convolutional neural network model to extract image information,and target tracking is performed under the framework of kernel correlation filtering algorithm.It is proved by comparison experiments that the online convolutional neural network model can be used to extract target features.Finally,a brief summary of the thesis is made,and the future work and innovations are prospected.