Quantitative Study On Drivers’ Vigilance Level Of Mountain Road Considering Human Factors And Environmental Factors

Traffic accidents on high-risk road sections in mountainous areas are related to various factors,such as drivers,environment,etc.When performing driving tasks,drivers usually face multiple pressures from the environment and their own psychology,and they need to constantly adjust their alertness to adapt to different road environments and avoid corresponding risks.Quantitative analysis of driver alertness and targeted warning based on comprehensive consideration of human and environmental factors are of great significance for improving the dynamic safety prevention and control level on high-risk road sections in mountainous areas.Therefore,in this study,a method for quantifying driver alertness on mountainous roads is proposed by conducting real-car experiments and analyzing and integrating human and environmental factors.Through analyzing the relevant concepts of driver’s psychological factors and summarizing the basic elements of existing alertness research,a driver alertness level is defined and a driver alertness level index system for mountainous roads is constructed,including 17 initial variables for quantifying alertness level,13 observation indicators,and 2 verification indicators.Twenty local drivers were recruited to conduct experiments on a 35 km section of the 129-county road in Guizhou Province,with rich linear and roadside scenes,and 18 sample sections were analyzed to extract 226 sets of alertness state samples.Information fusion and clustering methods were used to reduce dimensionality and group the observation indicators.A scorecard model based on logistic regression was constructed to convert the output probability into a 0-100 alertness score and achieve the quantification of driver alertness level on mountainous roads.The alertness level was rated into 5 levels,and the characteristics of drivers and environments for each level were summarized.The main conclusions are as follows:(1)Based on kernel principal component analysis,13 observation indicators were reduced to 4 fusion feature indicators with a cumulative contribution rate of 93.67% to fully extract the original data information.The sample was further divided into low alertness state samples(58)and high alertness state samples(168)based on K-means clustering algorithm.(2)The alertness level quantification model extracted 6 alertness level feature factors from 17 human and environmental factors,namely,driving experience,neuroticism,road section type,roadside safety level,roadside interference level,and distance to the uphill bend.The model had a KS value of 71.86% and an AUC value of0.907,showing good performance.(3)The characteristics of driver samples with different alertness level grades were summarized.The characteristics of drivers with low alertness level were middle score in Eysenck neuroticism,driving experience of 0-19 years,and environmental factors such as distance to uphill bends less than 80 m,roadside safety level of II and III,roadside interference level of I and II,and small and large radius curved slope sections.Reasonable safety suggestions were proposed for different road environments and human factors.