Intelligent System for Urban Road Driving Risk Identification Based on Machine Vision and Fuzzy Inference

To enhance driving safety and comfort, this study aims to investigate the level of driving risk under different road conditions in urban areas. The you only look once version five-deep simple online and real time tracking (YOLOv5-DeepSORT) model is employed to collect traffic data of forward vehicles, and novel methodologies are proposed to calculate the relative orientation, distance, and velocity between the forward vehicles and the main vehicle. Additionally, this study introduces an innovative function to determine the proximity of the forward vehicles to the main vehicle’s lane, along with a straightforward model for vehicle lane change recognition. By utilizing the relative orientation, distance, and velocity between forward vehicles and the main vehicle, information on forward vehicles’ lane changes, and the main vehicle’s velocity as inputs, an urban road driving risk identification model based on Mamdani fuzzy inference is constructed. The model was validated with real vehicle experiments that were conducted under various road conditions in the city of Zhenjiang. The results demonstrate that the proposed driving risk identification model achieves an impressive accuracy rate of 92.44% and exhibits good identification performance.