An urban traffic accident prediction model based on the full range of human, vehicle, road and environmental factors

Abstract

Traffic is a dynamic coupled system consisting of four factors: humans, vehicles, roads, and the environment. Safe and controllable external stimuli and the psychological condition of drivers guarantee the safety of vehicle driving, but due to the lack of quantitative means of assessing driver and environmental factors, accident analysis and prediction methods often lack scientificity. This reduces the effectiveness of traffic accident prediction and control. In this study, a full-factor urban traffic accident prediction framework is constructed using real-vehicle test data and open-source data to achieve the accurate prediction of urban traffic accidents. The real-vehicle test data include 1022 sets of experimental data obtained by 23 experimenters for more than 7 consecutive laps on 6 routes with large variability. The wide application of deep learning in computer vision and the popularity of physiological acquisition technology in daily life provide quantitative means for assessing driver and environmental factor variables, and the open-source data include traffic volume data from the drip platform, street view data and road design information. In addition, the analysis methods involving the elasticity coefficient and marginal benefit coefficient are used to analyze the potential connection between correlated variables and traffic safety. Based on this, all-factor traffic safety improvement measures are proposed, providing theoretical support and methodological support for the improvement of traffic safety on urban roads.