Research On Visual Target Tracking Algorithm Based On Correlation Filtering In Complex Scenes

Target tracking is an important research topic in the field of computer vision,and it has a wide range of applications in human real life,scientific and technological applications,military,national defense security and other fields.At this stage,target tracking technology has made great progress,and a large number of excellent tracking algorithms have emerged to cope with the urgent needs of complex scenes such as occlusion,lighting changes,target deformation and rotation,scale changes,and background interference in reality.Among them,the target tracking algorithm based on correlation filtering has significant effects in terms of accuracy and efficiency,but due to the influence of many complex scenes in reality,it still faces many challenges.In order to improve the tracking performance of related filtering algorithms in complex scenarios,this paper carries out the following research work:(1)Aiming at the problem that the single-feature kernel-related filter tracking algorithm cannot accurately describe the target and the anti-interference ability under complex scenes such as illumination and scale changes,deformation,and background mixing,a multi-scale correlation filter tracking algorithm based on feature fusion is proposed.First,in the position prediction stage,extract the HOG and color CN features,and adaptively weight the fusion features according to the response value of each feature to realize the prediction of the target position;secondly,in the scale prediction stage,train the scale filter on the predicted position of the previous frame to establish the scale The pyramid is used to adjust the scale of the target,and the scale corresponding to the maximum response value is obtained as the optimal scale of the target,which enhances the adaptability of the target scale.(2)In order to further improve the robustness of the correlation filtering tracking algorithm in complex scenes,a correlation filtering tracking algorithm based on depth features is proposed,and the deep feature training model with strong sample representation ability is used.First,analyze the comprehensive performance of each layer feature in the tracking scene,use the variance filtering method to extract the spatial and semantic depth features and input them into the related filter tracking model,and use fixed weights to fuse the response values of the features of each layer to achieve the accuracy of the target.Prediction;secondly,in order to prevent the algorithm from incorrectly updating the model when it encounters occlusion,which leads to tracking drift,a high-confidence model update strategy is proposed in the model update stage to calculate the response peak and average peak energy indicators to adaptively update the filter model to reduce the number of updates,Improve tracking accuracy.In the experimental part,the algorithm of this paper is tested on the OTB-2013 video sequence set with various interference attributes.Through quantitative and qualitative analysis,the tracking performance of the algorithm in this paper is compared with the traditional classic algorithm and the mainstream algorithm combined with deep learning in complex scenes.It shows that this paper The algorithm has significantly improved accuracy and success rate.