Safety effectiveness and performance of lane support systems for driving assistance and automation – Experimental test and logistic regression for rare events

As of 2022, lane-keeping assistance systems and other selected safety technologies will become mandatory in new European vehicles to increase safety for passengers, pedestrians and cyclists. Lane support systems (LSS) are based on advanced computer vision technologies and they are expected to give safety benefits in reducing Ran of Road and head-on crashes comparable to already consolidated physical countermeasures like rumble strips. Anyway, despite the assumed technology readiness, there is still much uncertainty regarding the needs of vision systems for “reading” the road and limited results are still available from in field testing. In such framework the paper presents an experimental test of LSS performance carried out in two lane rural roads with different geometric alignments and sections characterized by variable maintenance conditions for pavement and markings. LSS faults were detected in 2.6 % of the sections, running the roads in day light and dry pavement conditions. Logit models were developed to better understand road characteristics and conditions that can affect the system performance. The Firth penalized maximum-likelihood method was applied to estimate the logistic regression coefficients and standard errors to account for the rareness of the events. Results show that luminance coefficient of marking in diffuse lighting conditions (Qd) and horizontal curvature radius (1/R) are the main predictors of system fault. Based on the case study and test conditions, other marking characteristics, longitudinal cracking, verge width and running speed resulted not significant in explaining the probability of fault. Thresholds values for Qd and 1/R are suggested and remarks on road maintenance and design standards presented.