Research On Tracking Algorithm Based On Spatio-temporal Constraint Correlation Filtering Model

With the development of computer science and artificial intelligence,visual tracking technology has been widely used in video surveillance,navigation and positioning,aerospace driving and other fields.The essence of tracking is to continuously determine the position and size of the target in subsequent frames based on the target information in the first frame of the video.However,when the target has interference such as rapid motion,deformation,occlusion,etc.,the existing target tracking algorithm cannot accurately predict the position.These problems and in order to improve the performance of the tracker,the article is based on the spatio-temporal regularization filtering target tracking method to improve.The specific work is as follows:(1)First,in order to improve the tracking accuracy and anti-interference of the target,the depth feature and manual feature of the extracted target are selected as the feature to be tracked,and optimization is made based on these two features.The authenticity analysis method of the sample feature channel is introduced internally,and the contribution of different dimensional features to tracking is fully explored.The GMM model is also used to manage the sample set,and the historical frame samples and current frame information are added to jointly train the filter based on the spatio-temporal regularization model.The ADMM optimization algorithm is iteratively solved,and finally the response values of all features are added to obtain the maximum confidence score to determine the target position of the next frame.(2)Then,use manual feature weighting to improve the scale filter to estimate the scale,and improve the boundary effect caused by fast motion.Changing the spatial weights in the spatiotemporal regularization model and choosing the anti-Gaussian space model as the spatial regularization constraint can introduce more background information in the training,thereby making the content of the filter learning more comprehensive and distinguishing between the target and the background.(3)Finally,in order to enable the tracker to track for a long time,the tracking quality of each frame is evaluated to update the model.Introduce the response peak fluctuation amplitude and the maximum response score ratio between the historical frame and the current frame to determine whether the target is blocked or track deviated,and further filter pure samples to update the sample set and filter model online to keep the filter from being contaminated and achieve continuous tracking.In this paper,the algorithm is improved based on spatio-temporal regularization correlation filtering tracking algorithm.Experiments are compared with a variety of classic tracking algorithms in the OTB-100 data set.The results show that the proposed algorithm is more robust than the traditional target tracking algorithm when the target has background illumination interference,fast motion,deformation and occlusion.