Method for Identifying Motor Vehicle Traffic Violations Based on Improved YOLOv Network

Abstract

The use of traditional manual supervision means to deal with motor vehicle traffic safety violations can result in a large amount of wasted manpower and oversight problems. To assist road managers in better directing traffic order and managing traffic situations, the study proposes an improved target tracking network model. Simple online real-time tracking, deep correlation metrics, and cascading open-source computer vision libraries are combined to create a tracking model for motor vehicle traffic infraction recognition. Pursuant to the experimental findings, the Institute’s upgraded target recognition network model had accuracy and recall rates of 95.7% and 99.7%, respectively, with an accuracy rate of 16.6% higher than the model’s historical counterpart. The recognition accuracy of the constructed motor vehicle traffic violation recognition and tracking model regarding the three basic traffic violations was 98.2%, 98.7%, and 97.9%, respectively; the missed detection rate was 2.0%, 0.31%, and 2.1%, respectively; and the false detection rate was 0.17%, 0.31%, and 0%, respectively. It shows that the improved network model of the study is advanced and the motor vehicle traffic offence model has a good recognition rate and stable performance, which can assist traffic managers in their operations to a certain extent.