Research on Video Tracking Algorithm Based on Yolo Target Detection

Abstract

With the development of computer vision technology, target tracking is extensively used in traffic intelligence, security and many other realms. Deep learning is considered as an important supporting technology in the realms of video object tracking because of its excellent object modeling ability. The real-time target detection algorithm YOLOv3 has fast detection speed and good accuracy, but it has some defects, such as inaccurate boundary frame positioning and poor performance of multi-target tracking at high frame rates. In this paper, we propose a multi-target tracking algorithm based on YOLOv4 and SORT algorithm, which is based on deep learning neural network frame. First, YOLOv4 is used for target detection, and the detected pedestrian data is passed to SORT algorithm to realize multi-target tracking, and good performance can be achieved at high frame rate. By comparison, it can be concluded that the proposed multi-target tracking algorithm can improve the corresponding mAP value by about 6% and FPS value by 5s.