Research On Correlation Filtering Moving Target Tracking Method Combined With Spatiotemporal Context Information

As an important branch of computer vision basic research,moving target tracking technology is widely used in human-computer interaction,intelligent transportation,video detection and military fields,and has broad development prospects.The task of target tracking is to monitor the state of the target in subsequent videos based on the starting information of the target in the video sequence.Although the research of target tracking algorithms has made great progress in recent years,due to the interference of various complex factors during the movement of objects,how to design a stable and efficient target tracking method is still a very challenging problem,especially tracking frequently encountered problems with light transformation,occlusion,and model drift during the process.The main research contents of this article are as follows:(1)In view of the current problems of the target tracking algorithm,this paper proposes an improved spatiotemporal context target tracking algorithm.The algorithm first introduces color features that are robust to illumination changes,and after adaptive dimensionality reduction and grayscale features are used to describe the target and local context areas,and then calculate the context prior probability;Then based on the context prior probability to accelerate the calculation in the frequency domain to establish a spatial context model;Finally,the target confidence map is obtained by maximizing the likelihood function,and the position with the highest likelihood probability is the target position for the next frame.The experimental results show that,compared with the improved algorithm,the algorithm in this paper accurately predicts the target movement position in the tracking scene facing the occlusion challenge factor and the illumination change challenge factor,and achieves a more robust target tracking.The algorithm is not only applicable in multiple scenarios,but also runs fast,basically meeting the real-time requirements of the algorithm.(2)Aiming at the problem of poor target tracking stability and susceptibility to occlusion drift in complex environments,a correlation filtering target tracking method combined with spatiotemporal context information is proposed.The algorithm first extracts directional histogram features and color histogram features from the target and background areas,weights the fusion filter response of the two features,and establishes a correlation filter tracking model;Then uses the above features of the target,based on the bayesian framework to obtain space-time context-assisted model;Finally adaptively fuses the correlation filter response and the auxiliary model response,obtains the target estimated position by maximizing the response graph,and uses the scale estimation method to solve the target scale change problem,which effectively improves the tracking accuracy of the algorithm.A comprehensive experiment was carried out on the OTB-2013 open standard test set with a moving target tracking method based on correlation filtering compared.The results show that the average distance accuracy value and average overlap accuracy value of this algorithm are better than other algorithms,which can effectively alleviate the occurrence of tracking drift caused by occlusion,scale change,lighting and other factors.