Visual Tracking Via Kalman Filter Spatio-temporal Context Learning

In the rapid development of science and technology,all kinds of machines are moving toward the direction of intelligence,and computer vision has also been rapidly developed as a key discipline of smart devices.Target tracking,as a part of it,is currently applicable to smart monitoring,biometrics,and military surveying,etc.Target tracking primarily selects the proper tracking algorithm to track the stationary or moving target in the video frame.In this process,the background may change,and the target object itself may also undergo some deformation.How to deal with these changes is exactly the key of tracking algorithm.At present,there are many research institutions at home and abroad who are researching target tracking technology and have consistently emerged many excellent tracking algorithms.The spatio-temporal context tracking algorithm used the context around the target to predict the target position,so that the algorithm can respond to the changes in external scene and the partial occlusion of the target,improving the robustness of the algorithm.At the same time,fast Fourier transform is used to make the algorithm speed performance good.However,when the target is completely obstructed,the algorithm will appear to track the loss of the target.When the target reappears,the algorithm can't detect the target because it has no detection function.In order to solve this shortcoming of the algorithm,this paper presents a visual tracking via kalman filter spatio-temporal context learning.The main tracking process of this algorithm: First,calculate the RGB color histogram of the current frame and the template image.The RGB color histogram contains the color characteristics of the moving object,and this basically remains unchanged during the movement of the object,so relatively accurate results can be obtained relatively.Then,using the RGB color histograms of the two samples to calculate the Bhattacharyya between them,the similarity measure standard was used to judge the occurrence of occlusion.If the Bhattacharyya is less than the threshold selected in advance and the target and the template are not similar,the global occlusion of the target is determined and Kalman filtering is used to predict the target tracking result;otherwise,the STC algorithm is used to keep track of the target position.A large number of experiments show that the algorithm improves the tracking success rate of the test sequence to more than 80%,and the center position error is also much smaller than the original STC algorithm and some other advanced algorithms.At the same time,the operating speed of the KSTC algorithm is only reduced by a small amount,retaining the advantage of the high rate of the STC algorithm and having certain efficiency and real-time performance.The KSTC tracking algorithm proposed in this paper not only retains the advantages of the STC algorithm in dealing with complex scenes such as lighting and shadow changes,sudden movements,and scale changes,but also has good anti-occlusion capabilities,especially in the case of global occlusion.STC algorithm accuracy and robustness.