Research On Target Tracking Algorithm Based On Multi-feature Fusion And Redetection Mechanism

As an important branch in the field of computer vision,target tracking technology is one of the important means of information transmission between human and computer,which is crucial to the development of artificial intelligence.Target tracking technology is widely used,the scene is complex and diverse,and the target movement is irregular,which poses a great challenge to the actual tracking.The method in this paper is mainly applied to mobile camera equipment,which has a big limitation on its calculation.How to improve the accuracy of the target tracking algorithm while reducing the amount of computation,so that it can run stably in intelligent equipment,and has a high anti-interference ability is the focus of this paper.This paper studies the work as follows:(1)a multi-feature fusion correlation filtering target tracking algorithm is proposed.The algorithm describes the target information in the image by the color feature and the shape feature.Color Naming,Color histogram and the histogram of Oriented Gradient method for target feature extraction and modeling respectively,using the filter to the target template mutual correlation calculation with the next frame image generated three relevant response matrix,and through the scale after normalization,set up the super parameters weighting and three relevant response matrix of weighted fusion to generate the final response template.Experimental results show that the algorithm is effective in overcoming the interference of scene information.The algorithm was 3.6% higher than other algorithms in terms of success rate and 3.4% higher than other algorithms in terms of accuracy,with an average frame rate of 58 frames per second.and meets the real-time requirement.(2)a redetection mechanism of tracking algorithm is proposed to overcome the total occlusion or out of view of the target.The algorithm consists of three main modules,namely,unreliable evaluation,redetection module and reliability evaluation.When the tracking effect is poor or the tracking fails in the tracking process,the peak value of the response matrix will be greatly reduced and lower than the unreliable evaluation threshold.When the peak value of five consecutive frames of images is lower than the unreliable evaluation threshold value,the redetection module starts,and the particle filter method is used to search and update the target position.When a new target position is found,the reliability evaluation calculates the correlation between the new target and the target template to obtain the peak value of the response matrix,which is greater than the reliability evaluation threshold,and then passes the target position to the target tracker.The experimental results show that this method can deal with the feature full occlusion and the field of view,and ensure the long-term and stable tracking effect of the tracker.Compared with the algorithm in chapter 3,the accuracy index is improved by 2%,and the success rate index is improved by 2.2%.The average frame rate of the algorithm in this chapter is 46 frames per second,which meets the real-time requirements.(3)In order to facilitate the detection of the effect of the algorithm in practical application,a real-time target tracking system based on user graphical interface design is built on Matlab.The system has five target tracking function modules,which can synchronously output the initial target image,the current target position,the current target image,the target grayscale map and the target response map of the target tracking algorithm.By using these modules,the effect of the target tracking algorithm in actual use can be effectively monitored.