Lane departure on combined curves: driver heterogeneity, centrifugal risk, and crash prevention

Lane departure crashes were more likely to result in severe traffic crashes. This objective of this study was to explore the effects of drivers' characteristics on the asymmetric lane departure behaviour in combined curves. A high degree of freedom driving simulator was used to collected vehicle operation data in a mountainous freeway. Totally 948 lane departure events were collected. The lane departure behaviour was classified into two categories based on centrifugal force direction: In the Direction of Centrifugal Force (IDCF) and Against the Direction of Centrifugal Force (ADCF). The average speed, maximum lateral departure and departure duration distance were analyzed. A Multivariate Adaptive Regression Splines (MARS) model was used to analyze nonlinear relationships between driver characteristics and maximum lane departure. It found that: (1) Sag-curves and crest-curves had higher average lane departure frequencies compared to downslope- and upslope-curves. (2) IDCF events showed greater departure (0.83 m > 0.41 m) and longer departure duration distance (70.07 m > 58.60 m) than ADCF events. (3) The years of driving, daily driving distance, driving experience, road expert type and departure duration distance had significantly influence on departure severity. Also, driver characteristics and lane departure duration distance had an interaction effect on departure severity. (4) Speed and road geometry design could jointly influence the degree of lane departure, e.g., speed significantly affects lane departure behaviour on downslope-curves and upslope-curves in the IDCF scenario. (5) The lane departure duration distance threshold (35.00-110.00 m), and speed threshold (86.77-108.63 km/h) were proposed for the lane departure process. The study provided a basis for the optimisation of ADAS systems, the development of targeted driver training for high-risk drivers, and the optimisation of visual guidance design for combined curves.