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

Area of Science:

• Road safety engineering
• Transportation psychology
• Computer vision

Background:

• Mountainous highways face high accident rates due to challenging topography and road design.
• Existing research often overlooks the discrepancy between actual road geometry and drivers' visual perception.
• Driver behavior is significantly influenced by perceived visual information, not just objective road features.

Purpose of the Study:

• To quantify drivers' visual perception of road geometry on two-lane mountain highways.
• To identify key visual perception parameters differentiating accident-prone from accident-free locations.
• To develop a model for predicting accident-prone sites based on visual perception.

Main Methods:

• Development of a driver's visual lane model to quantify visual perception.
• Projection of the visual lane model onto horizontal and vertical planes to extract shape parameters (length, curvature).
• Conducting real vehicle driving tests on mountain highways and analyzing differences in visual perception between accident spots and safe locations using a probabilistic neural network (PNN).

Main Results:

• Nine shape parameters of the visual lane model showed significant differences between accident-prone and accident-free locations.
• A probabilistic neural network (PNN) was successfully developed to identify accident-prone locations.
• Quantification of visual perception provides a new metric for assessing road safety.

Conclusions:

• Driver's visual perception is a critical factor in mountain highway safety, distinct from actual road geometry.
• The developed visual lane model and PNN offer a novel approach for identifying and mitigating accident risks on mountain roads.
• Findings can inform road design, reconstruction, and advance autonomous driving system development.