Research On Road Vehicle Detection And Tracking Algorithm Based On Deep Learning

With the wide application of artificial intelligence technology in the field of traffic vehicles,autonomous driving systems and intelligent traffic management methods can reduce the number of traffic accidents and improve urban traffic congestion.Among them,vehicle detection and tracking is an important part of realizing automatic driving and intelligent transportation,but due to the increasingly complex traffic environment,traditional object detection and tracking technology is difficult to meet the performance requirements of vehicle research.In recent years,deep learning has rapidly developed into the most popular application method in the field of machine vision,and has unique advantages in vehicle detection and tracking.Therefore,this paper carries out the research on vehicle detection and tracking algorithms on the basis of the theory of deep learning,and the specific research content is as follows:(1)In the vehicle detection task,the YOLOv5 detection algorithm is effectively improved in view of the problems of low vehicle detection accuracy and limited realtime performance in the vehicle target detection stage.Firstly,two improved lightweight vehicle detection models,YOLOv5-Ghost and YOLOv5-Rep VGG,are proposed,and then the attention mechanism module is applied to the neck network of YOLOv5,and the loss function of the model is improved to improve the accuracy and speed of border regression.In order to verify the effectiveness of the proposed algorithm,the improved YOLOv5 vehicle detection algorithm is experimentally evaluated on a self-made dataset,and the results show that the algorithm achieves a good balance in speed and accuracy.(2)In the vehicle tracking task,the tracking algorithm DeepSORT is improved for the problems of vehicle occlusion and low tracking accuracy in the vehicle target tracking stage.Firstly,the original feature extraction module in DeepSORT is improved to improve the quality of vehicle appearance features,and then the motion feature measurement method in the data association stage is optimized to improve the matching accuracy in the tracking process.In this paper,The algorithm that combines the improved YOLOv5 and DeepSORT is evaluated on a dataset for tracking experiments,and the results show that the algorithm can better realize vehicle tracking in the actual road environment.