Design And Implementation Of Target Tracking Based On Kernelized Correlation Filter

Nowadays,moving target tracking is one of the most challenging tasks in the field of computer vision,which is applied to human-computer interaction,intelligent traffic monitoring and robot.With the rapid development of target tracking and recognition technology and the continuous improvement of security intelligent monitoring,The intelligent requirements of intelligent monitoring,terrain navigation,video intelligent labeling and other applications are increasingly high.Intelligent monitoring system refers to the computer can automatically process the image transmitted by the monitoring equipment,and conduct corresponding behavior analysis on the movement and attitude of moving objects,etc.,which mainly involves artificial intelligence,pattern recognition and big data processing,when the monitoring system is in the case of non-human interference and control.Target tracking is an indispensable part of intelligent monitoring system.Its performance determines the strength and performance of the monitoring system.Despite to numerous attempts,it is still a challenging task to overcome the disturbance factors in the tracking of moving targets,such as scale change,occlusion,sudden movement,light change and rotation inside and outside the plane.This paper aims at the problems existing in the current target tracking and combines with some common tracking algorithms in the industry for in-depth research,the specific results are as follows:1?An enhanced kernelized correlation filter algorithm with multi-feature fusion is proposed.The algorithm can effectively reduce the influence of illumination change,occlusion,size change and other factors in the tracking process of moving target.The algorithm firstly extracts the image's directional gradient histogram feature information,color name information,local binary texture information and Canny edge information to fuse them into a group of multi-feature information for target tracking.Then the multi-scale search method is adopted to the scale change of the target region.Finally,the maximum output response score of the previous frame and the current frame is compared to determine whether the occlusion and other conditions of the target exist.If occlusion and other conditions exist,the model parameters are updated again.Experimental results show that this algorithm has higher accuracy and robust than the traditional target detection algorithm when the target tracking process is affected by fast motion,occlusion and scale.2?A visual tracking algorithm based on kernelized correlation filter is proposed.Because the target is selected as the template training sample in the first frame,it is difficult to solve the problems of occlusion and target appearance deformation.Because of the single sample,it don't have the ability to recover from complex template.So the algorithm based on clustering method to design the framework of template matching,can more accurately track the target through this framework,this paper also puts forward a random update template matching strategy,determine the nuclear correlation filtering factor of learning,so as to realize the adaptive updating model learning factors.Then,based on the clustering method,the algorithm designs a multi-template matching framework.At the same time,this paper proposes a random updating multi-template matching strategy,It determines the learning factor of kernel correlation filtering and realizes the adaptive updating model of learning factor.By adapting learning factor and multi-template matching model,the algorithm has strong adaptability to partial occlusion,illumination and target scale change.3?The design and realization of target tracking system are proposed.This system is based on the QT platform,combined with OpenCV and other visual libraries,to achieve the monitoring scene of different targets tracking.System functions include:(1)target detection and tracking function:this software can track moving targets of specified video files;(2)target edge detection function:this software can be used as edge detection to provide detection of specified files;(3)video playback function:this software can be used as a player to provide the playback of video files in various formats;(4)video recording function:this software can conduct qualitative recording of specified video files according to user requirements.