Study On Moving Human Body Detection Technology Based On TLD Algorithm

The birth of the computer opened the door of the intelligent age for human beings.With the rapid development of computer technology,more and more technical achievements are widely used in human daily life and work,and will continue to serve human beings and benefit mankind.In the field of computer vision,moving target detection and tracking has been a major focus of scholars and experts,how to detect the moving object quickly and accurately using the image frame sequence,while effectively the trajectory of the moving target,it is a very rich application value and challenges.This paper focuses on the TLD target detection and tracking algorithm,and on the basis of its basic principle,Meanshift algorithm is added to optimize.The TLD algorithm combines the target detection task with the target tracking task,works together with the moving target tracking,and adds the tracking failure self detection function,which can re initialize the tracking template when the tracking fails.In the detector,3 layers filter is used to detect the closest target template,and the accuracy of the detector is improved by using the statistical machine learning model,fern integrated filter and nearest neighbor filter.However,the disadvantage is that the time overhead of the tracker is large,and the tracking speed is affected in the video frame sequence.The Meanshift algorithm as the target tracking field and basic algorithm,using the mean shift iteration to update the position of the center point,adding kernel function makes the different influence on the offset value of mean shift is different,the weight of adding different sample points on the target in the heart's contribution is different,therefore shows good effect in target tracking in.But the disadvantage of Meanshift is that the target template can not be updated dynamically,and the tracking effect is decreased in complex scenes.In this paper,the basic principle of detection and tracking algorithm is studied,and the TLD algorithm tracking module is used to replace the slow moving optical flow method.Instead,the Meanshift algorithm with less time cost and fast convergence is optimized,and the tracking efficiency of TLD algorithm is optimized.At the same time,the traditional Meanshift algorithm only on the first frame image tracking initialization template tracking target transform does not update the tracking template,aiming at the shortage of tracking template cannot be updated,in the improved algorithm with an updated link tracking template,using the detected templates in the tracking module to update the template tracking failure tracking,adaptive motion the target with the displacement changes.The experiment results show that the tracking time cost is reduced,and the tracking effect is improved to some extent.