Effects of a bicyclist detection system on police-reported bicycle crashes

April 2023

Research question/objective: Automatic emergency braking (AEB) is effective at preventing vehicle-to-vehicle rear-end crashes and pedestrian crashes. Subaru's driver assistance system that includes AEB, called EyeSight, could detect bicycles in parallel configurations in the United States in its first and second generations, and added bicyclist detection in perpendicular configurations in its third generation. The purpose of this study was to evaluate whether the first and second generations of EyeSight reduced bicycle crashes in the real world. Methods and data sources: The presence or absence of EyeSight was identified through Vehicle Identification Numbers for model year 2013–2020 Subaru models where the system was optional. All bicycle crashes and single-vehicle single-bicyclist crashes with parallel and perpendicular configurations involving these vehicles were extracted from the police-reported crash databases of 16 U.S. states during calendar years 2014–2020. The association of EyeSight with bicycle crash rates per insured vehicle year was examined with negative binomial regression controlling for calendar year, state, vehicle model year and series, and driver age group and gender. Quasi-induced exposure analyses using logistic regression compared involvement in a bicycle crash to the nonsensitive crash types of being rear-end struck or side-struck, using the same covariates as the negative binomial regression models. These analyses included crash data from 14 states where rear-end-struck and side-struck vehicles could be identified. Results: Study vehicles were involved in 856 bicycle crashes, of which 283 had parallel configurations and 387 had perpendicular configurations. EyeSight was associated with a statistically significant 29% reduction in parallel crash rates per insured vehicle year (Rate ratio [RR], 0.71; 95% confidence interval [CI], 0.53–0.96, p = 0.03), and nonsignificant reductions of 5% in perpendicular crash rates (RR, 0.95; 95% CI, 0.74–1.21, p = 0.66) and 9% in overall bicycle crash rates (RR, 0.91; 95% CI, 0.77–1.08, p = 0.28). Effects of similar magnitudes were seen in the quasi-induced exposure analyses. Discussion and limitations: An early version of EyeSight reduced bicycle crashes in the parallel configurations it was designed to detect but did not have much effect on bicycle crashes overall. Crash configuration was identified by bicyclist and vehicle direction of travel when they were available. In states where direction of travel was unavailable, bicyclist precrash actions of cycling along the roadway with or against traffic and crossing were used as proxies for parallel and perpendicular configurations, respectively. The actual configurations of crashes in these states were unknown. Conclusions: Although it is promising that an initial bicyclist detection system prevented crashes in parallel configurations, a minority of bicycle crashes are of this type. AEB systems will need to increase functionality and detect perpendicular crash configurations to meaningfully reduce bicycle crashes.