Grid Positioning Guidance Based On Improved KCF Algorithm For Object Tracking

With the rapid development of computer vision related fields,target tracking,as an important branch of computer vision,has been sought after by various industries,which also makes the scenes encountered by target tracking algorithms in the tracking process more and more complicated.In the tracking process,occlusion,deformation,complex background,multi-scale and blurry issues have yet to be resolved.The KCF tracking algorithm based on kernel correlation filtering came into being.Because of its good temporal robustness,it has received extensive attention since its inception.In this paper,based on the KCF target tracking algorithm,a series of improved methods are proposed to address the shortcomings of its spatial robustness.In order to obtain a target tracking algorithm that meets both temporal robustness and spatial robustness,this paper proposes a new spatial grid positioning(GPG)tracking model.The GPG algorithm uses an average peak-to-peak correlation energy the rate of change(APCH)determines whether the tracking of the target in the current frame is reliable.If it is reliable,continue to use KCF for tracking.Otherwise,the spatial grid-oriented positioning model is used to track the tracking target.Therefore,the GPG model can be used to solve the target tracking process.The target occlusion,target scale change,background blur and target drift bring about the problem of poor spatial robustness of the tracking algorithm.This paper uses the following three methods to solve the problem of KCF target tracking algorithm in the tracking process.(1)Target frame positioning: Using grid feature fusion to generate a new preselected frame makes the positioning of the target tracking process more accurate,which can effectively prevent the target frame from shifting,ending the problem of poor spatial robustness of the KCF algorithm.(2)Update strategy: Using APCH instead of APCE can make the entire evaluation standard more refined.Using the derivative of the average correlation peak energy makes the tracking response smoother,easier to find the catastrophic response,find the problematic target picture,and locate Go to the problem frame,and then integrate the spatial grid positioning to reposition and track.(3)Target scale change: By mapping the pre-selected frame in the initial frame picture in the tracking process to expand the area,the problem that the target frame cannot follow the target suddenly becomes larger and larger during the tracking process is solved.The GPG algorithm better reflects the KCF algorithm's tracking reliability of the current frame during the tracking process through the APCH,and optimizes the spatial robustness of the tracking frame through the grid positioning model.Through the combination of KCF and deep learning,a new type of spatial grid target tracking algorithm GPG is proposed and analyzed experimentally.The experimental results in the target tracking data set OTB-50 show that the GPG tracking algorithm can effectively solve the tracking process.Related issues,and has good target tracking performance.