Research And Implementation On Real-time Detection And Tracking Algorithm To Vehicles And Pedestrians Based On Deep Learning

Computer vision is a comprehensive subject in computer science,covering knowledge in many fields.By using computers and its video equipment to replace the brain and visual system of human,the computer can independently complete the analysis and understanding of images or videos.Target detection and tracking is an extremely important research direction in the field of computer vision,and it also plays an irreplaceable role in the development of artificial intelligence.With the continuous innovation of computer-related technologies and the continuous in-depth research of artificial intelligence,target detection and tracking technology has become more and more widely used in multiple fields including smart transportation,intelligent security,military and medical.This paper proposes a detection and tracking algorithm to vehicles and pedestrians based on deep learning.It takes vehicles and pedestrians on traffic roads as the research goal,adopts the improved YOLOv3 algorithm for rapid detection,and combines the multi-scale matching model with the Hungarian algorithm to enable more accurate tracking of moving vehicles and pedestrians.The main tasks of this article are as follows:(1)Research on detection algorithm to vehicles and pedestrians based on improved YOLOv3.The current mainstream two types of deep learning-based target detection algorithms are studied in the article,and the YOLOv3 algorithm is improved for the problem that the YOLOv3 algorithm does not detect well in target-intensive scenarios and has poor recognition of small targets.The network structure of Darknet-53 is adjusted,a feature scale is added,and the value of Anchors is reset according to the data set used.The improved YOLOv3 algorithm is applied to detection of vehicles and pedestrians on traffic roads,which effectively improves the detection performance of vehicles and pedestrians of different scales.(2)Research on real-time tracking algorithms to vehicles and pedestrians based on Kalman filtering.The algorithm first uses improved YOLOv3 algorithm to detect vehicles and pedestrians in the real-time video sequence,and then uses the fusion model of Markov distance,cosine distance and spatial distance to associate the detected target object with its trajectory,and then completes the matching by the Hungarian algorithm,finally the tracking algorithm based on the Kalman filter is used to modify the tracking prediction result according to the matching result for fast,accurate and continuous tracking of vehicles and pedestrians.In order to further verify the comprehensive effect of the tracking algorithm,this paper designs an intelligent traffic monitoring system based on crawler robot to detect and track pedestrians and vehicles on the actual road,and realize real-time analysis of road conditions..