B. Kutela, M.P. Mihayo, H.M.M.B. Khalaf, E. Kidando, A.E. Kitali

Pages: 103-118

Abstract
Lighting condition is an extensively adopted measure to improve visibility for vulnerable road users (VRUs). Despite the efforts to assess its influence on VRU visibility, previous studies have utilized simulations or staged VRUs, which may not represent the actual crash scenarios. Further, most studies singled out the lighting condition without interacting with other factors. Therefore, this study utilized crash data from the Ohio Department of Public Safety database collected between 2017 and 2022 to evaluate the influence of lighting conditions on VRU visibility. Crash narratives were used to extract crashes involving VRUs where drivers stated that they did not see them before a crash. Comparison crashes that occurred within 250 feet were extracted. Bayesian Networks (BNs) and Text Mining approaches were then applied. As expected, BNs results revealed that drivers were likelier to state that they did not see VRUs during dark, unlighted conditions. The combination of the variables indicated that drivers were more likely to report that they did not see VRU when it was dark and either VRU was standing, the driver was slowing down, the crash occurred at other locations, or the crash involved a senior driver. Further, text mining indicated additional details regarding senior drivers, driver actions, severity, and location of the crashes, among others, which would not be easily explored using traditional approaches. The study findings have the potential to inform targeted safety measures to reduce VRU-related crashes resulting from poor visibility, thereby enhancing road safety for all road users.
Keywords: vulnerable road user; VRU invisibility; Bayesian networks; text network