Research On Target Tracking Algorithm Based On Correlation Filter

With the development of computer technology,more and more people use computers to process video information,and intelligent video processing has become more and more important.Video target tracking is part of intelligent video processing and has a diverse range of applications in video surveillance,video navigation,intelligent transportation,human-computer interaction,weapon guidance,and medical image processing.In recent years,video target tracking technology has been continuously improved,but still faces many problems in practical application scenarios,such as target occlusion,attitude change,fast motion,rotation,etc.It is a challenging problem for this research on target tracking algorithm.The related filtering target tracking method has faster tracking speed and higher tracking accuracy.This paper studies the correlation filtering target tracking algorithm for the problems encountered in target tracking in complex scenes.The main work is as follows:(1)When the appearance changes in the complex scene,such as occlusion,deformation and illumination changes,this paper proposes a spatio-temporal context algorithm based on color features and it is based on the spatio-temporal context algorithm(STC).Since the existing target tracking methods mostly ignore the background information around the target,we extract the background information of the target's local context area.The spatio-temporal context algorithm uses the simplest gray feature to extract the features of the target.The low-level visual features are often too simple,ignoring the target's color information,texture information,structural information,etc.We weightedly merge the color features and gray features to describe the feature of a target.The spatial context model is constructed,and the spatial context model is used to update the spatio-temporal context model of the next frame and calculate the confidence map,with the maximum value in the confidence map as the estimated target location.(2)In order to solve the situation that the tracker updated errors and model drifted when updating the model,a scale-adaptive model update scheme is proposed.Since the target may vary in scale in each frame,we average the scales of the targets in the previous frames,update the scale of the next frame according to the estimated scales,and use the updated scales for the model update.The estimate is more accurate.In addition,increase the frequency of the number of video frames selected when estimating the scale,thereby improving the accuracy of the tracking.(3)Aiming at the problem that the target changes in the complex scene caused by the scale change,a target tracking algorithm based on position and scale estimation is proposed.The tracker of the algorithm consists of two correlation filters that estimate the target motion trajectory and target scales.Combine the estimated positions and scales to update the position and scale models.