Research On Target Detection And Tracking Algorithm Based On Video Image

The main purpose of computer vision target tracking problem is to predict the size and position of the target in the following frames on the premise that the size and position of the target in the first frame of the target tracking video sequence are known.Target tracking has always been one of the hot issues in the field of computer vision,because it has a wider application in the fields of public security monitoring,UAV,traffic monitoring,urban intelligent interaction system and so on.But at the same time,there are many challenges in the research of this algorithm,such as manual frame selection,target shape change,background illumination change,occlusion,motion blur and so on.How to deal with and solve these problems efficiently is a hot research direction at this stage,and also the main research task of this paper.In this paper,aiming at many problems of computer vision target tracking,we choose the target tracking algorithm based on correlation filtering and improve it according to its shortcomings and task needs,and at the same time,we test it by simulation,so that the algorithm can achieve better results.The main research contents are as follows:First of all,in view of the shortcomings of traditional target tracking algorithm,which relies on manual selection of target in the initial frame of video sequence,the idea of computer vision target detection is introduced before the tracking algorithm,and the target detection task is realized by using the series convolution neural network target detection algorithm.At the same time,because of the poor adaptability of the traditional convolution neural network target detection algorithm to the shape of the target,using the idea of deformation convolution to improve the traditional convolution neural network target detection algorithm,improve the feature extraction ability of the target detection algorithm,and then improve the accuracy of the system target detection,and realize the function of the system automatically detecting the target to be tracked.Secondly,the target tracking algorithm based on correlation filtering is adopted in the part of target tracking.In the traditional correlation filter target tracking algorithm,the feature demand is large and the description is poor,while the depth feature has strong description.The training of correlation filter is improved by using the depth feature extracted from the convl layer of the depth neural network VGG-M and the third convolution layer of resnet18,which makes up for the disadvantage of the poor description of the traditional correlation filter algorithm,making the correlation filter objective The accuracy of the target tracking algorithm is significantly improved.After that,the algorithm in this paper is tested by experiment.The experimental results show that the algorithm can complete the expected tasks,improve the accuracy of computer vision target detection and tracking algorithm,and make up for the shortcomings of traditional target detection and tracking algorithm.Finally,the work of this paper is summarized,and the research is prospected.