Research On Target Tracking Algorithm Based On Correlation Filtering Under Complex Scenario

With the continuous development and progress of science and technology,the demand for intelligent monitoring in human actual life is also increasing,and more and more attention has been paid to target tracking technology.In this thesis,we study the background clutter,target deformation,rotation and occlusion problems in target tracking in complex tracking scenarios.Based on the relevant filter tracking algorithm,we use the method of foreground perception,multiple complementary features and occlusion evaluation.The target tracking algorithm has been improved with the relocation method.The main research contents of the thesis are as follows:1.In order to solve the problems of background clutter,target deformation and rotation when the accuracy and robustness of the relevant filtering algorithm in the tracking scene,design an improved relevant filtering tracking algorithm based on foreground perception and fusion of multiple features.Use the pre-defined binary mask matrix to perceive the foreground target in the image to obtain the filter template with foreground perception capability;use the time-series consistency model to restrict the change of the filter template of the adjacent frame image to solve the model quality caused by the filter template jump the problem of decline.Analyze the advantages and disadvantages of HOG,CN and LBP features for target representation in complex scenes,design an improved multi-feature adaptive fusion strategy to improve the adaptability of tracking algorithms to complex scenes,and solve the problem of tracking algorithm performance degradation caused by target deformation and rotation to improve the adaptability of tracking algorithms to complex scenes.2.Aiming at the problem of being occluded by other objects during target tracking,an improved filtering tracking algorithm based on occlusion evaluation and relocation is proposed.The degree of occlusion is evaluated by calculating the peak side lobe ratio and the average peak correlation energy for the correlation response graph.When the target is evaluated as severe occlusion,the method of dynamically increasing the search range according to the occlusion evaluation is used to relocate the target.According to the evaluation of occlusion degree,an adaptive learning rate is used to update the improved correlation filtering algorithm model.3.Test the correlation filtering algorithm based on foreground perception and multi-feature fusion by selecting a sequence containing scenes with background clutter,deformation,and target scale changes from the standard data set and analyze the tracking results,proving that the proposed algorithm is used in complex scenes effectiveness.Using the image sequence selected from the OTB standard data set containing the target occlusion and out-of-view scenes,the tracking algorithm based on occlusion evaluation and relocation is experimentally tested and performance analyzed,and compared with other 7 target tracking algorithms based on correlation filtering to prove the effectiveness of the proposed algorithm in tracking targets in complex scenes.