Quantifying drivers' visual perception to analyze accident-prone locations on two-lane mountain highways

Owing to constrained topography and road geometry, mountainous highways are subjected to frequent traffic accidents, and these crashes have relatively high mortality rates. In middle and high mountains, most roads are two-lane highways. Most two-lane mountain highways are located in rural areas in China, where traffic volume is relatively small; namely, traffic accidents are mainly related to the design of roads, rather than the impact of traffic flow. Previous studies primarily focused on the relationship between actual road geometry and traffic safety. However, some scholars put forward that there was a significant discrepancy between actual and visual perceived information. Drivers greatly depend on what they perceived by their vision to determine driving behavior. Thus, in this paper drivers' visual lane model was established to quantify drivers' visual perception. To further explore drivers' perception of horizontal and vertical alignments, the visual lane model was projected onto horizontal and vertical planes in drivers' vision respectively. The length and curvature of the visual curve were extracted as shape parameters of drivers' visual lane models. Real vehicle driving tests were conducted on typical two-lane mountain highway sections of G318 in Tibet, China. Then the differences of visual perception at black spots and accident-free locations were analyzed and compared. In horizontal and vertical projections of visual lane model, there were 9 shape parameters have significant differences between accident-prone and accident-free locations. A probabilistic neural network (PNN) was formed to identify accident-prone locations on two-lane mountain highways. This study will lay a foundation for the improvement of traffic safety on mountain highways based on the quantification of drivers' visual perception, during the phase of both road design and reconstruction, and can also make a contribution to the automatic driving technique.