A Research On Multi-Feature Fusion Based Kernel Correlation Filtering Target Tracking Algorithm

Target tracking,as the core research content in the field of computer vision,has vital applications in many scientific and technological fields such as intelligent monitoring,unmanned driving,and human-computer interaction.The tracking method based on correlation filtering converts target tracking from time domain calculation to frequency domain calculation through Fourier change,which improves the accuracy and precision of target tracking.However,target tracking will be affected by factors such as deformation,occlusion,and illumination in a complex tracking environment.Traditional target tracking algorithms are difficult to meet the tracking requirements.Aiming at the problem of target tracking in complex scenarios,this paper analyzes the reasons that affect the accuracy and success rate of target tracking.Under the premise of ensuring the real-time performance of the algorithm,two improvements are made to the traditional algorithm in terms of model features and model update mechanism: A multifeature fusion algorithm;and a target tracking algorithm with adaptive model update to improve the accuracy and robustness of the relevant filtering target tracking algorithm.The main contents of this paper are as follows:First,a multi-feature fusion kernel correlation filtering algorithm is proposed to solve the problem that traditional target tracking algorithms mostly use a single feature for calculation,and cannot accurately track the moving target under the conditions of size change,occlusion,and image blur.According to the advantages of each feature in different environments,the output response value of different features is calculated,and the Bart Charia coefficient is used to weight the different features to complete the feature fusion of the decision-making layer,thereby improving the robustness of the algorithm.Then,the model adaptive update strategy is proposed to solve the model drift problem.The method is to set a threshold for the maximum response value of tracking,if the tracking is successful,update the model;if the tracking fails,stop updating the model;if it is judged to be affected,use the detection module to detect the target in the global scope of the image.Ensure that the model can be updated in a timely manner while suppressing model drift.Finally,this paper uses the OTB50 data set to fully experimentally verify and compare the proposed algorithm in a variety of different situations,and evaluate the tracking accuracy and tracking success rate of the algorithm in this paper.Experimental results show that the algorithm in this paper has a significant improvement in the robustness of target tracking.