Research On Video Object Tracking Algorithm Based On Correlation Filters

In recent years,with the continuous development of smart city and skynet systems in China,video target tracking technology has been widely used,such as intelligent monitoring in smart cities,face recognition and target tracking in skynet systems,etc.It also puts forward higher requirements for target tracking technology.Video target tracking has always been a hot topic in the field of computer vision.Since David SBolme first applied correlation filtering in the field of video target tracking in 2010,many scholars have conducted in-depth research on video target tracking algorithms based on correlation filtering in order to improve the accuracy of video target tracking.This thesis proposes a filter optimization scheme for the low target recognition success rate and target loss caused by scale changes in conventional video target tracking algorithms.By improving the adaptive strategy of the filter,the use of two evaluation indicators,maximum response value and average relative peak value,to intervene in the update of the filter model.The filter model is updated only when the tracking result meets the update conditions,which reduces the filter sampling frequency and prevents the filter from performing invalid sampling to affect the sample accuracy.In addition,a scale filter is designed to analyze the scale of the tracking target,and the recognition accuracy is adaptively adjusted according to the sampling size,to avoid the drift of the target tracking when the sampling accuracy is low,and to improve the target recognition rate.In order to solve the problem that the target is blocked and suddenly disappeared in the video target tracking,and to ensure the validity of the training samples of the training set,this thesis designed two anti-occlusion algorithms.The first method is the regional color histogram occlusion judgment algorithm.First draw the regional color histogram to form the histogram group value.By comparing the current target color histogram group and the target area historical average target color histogram group,the anti-occlusion algorithm determines the target is blocked and forms a filter update.,Tracker selection,search strategy and other judgments,and then use this basis to properly intervene in the tracker to achieve tracking purposes.The second method is an image segmentation occlusion evaluation algorithm.This method is mainly to extract the outline of the image,and then integrate the image segmentation algorithm and the target tracking algorithm to obtain the judgment basis of the occlusion of the tracked target,and then affect the retrieval and update behavior of the fused related filter tracker.The purpose of accurate tracking.The above two anti-occlusion algorithms can ensure the effectiveness of the training set training samples when the target is blocked or suddenly disappears during the video tracking process,which improves the accuracy of the training set training samples and the recognition success rate of the relevant filtering algorithm.