Assessing the impact of AR-HUD warning systems on cyclists' crash risk in connected environments: incorporating crash...

In connected environments, Head-Up Display (HUD) and Augmented Reality Head-Up Display (AR-HUD) alert drivers within their line of sight, yet their effects on crash risk in cyclist occupying lane scenarios remain unclear. This study aims to quantify the impact of different warning systems on the "probability-severity" integrated crash risk of cyclists and explore potential crash risk characteristics across different driver groups. Using a driving simulation platform, the cyclist occupying road scenario and three warning systems (Baseline/HUD/AR-HUD) were constructed in a connected environment. Driving behavior data from 38 participants were collected. Based on the research objective and conflict mechanism, DRAC was selected as the primary indicator to develop an Extreme Value Theory (EVT)-based crash probability model. In addition, a Delta-V-based crash severity probability model was established, and crash probability was further integrated with conditional injury severity to derive standardized expected-risk estimates for severe and non-severe crashes. The results showed that the DRAC-based EVT model reasonably characterized extreme conflict risk under the three warning conditions. Compared with the Baseline, both HUD and AR-HUD were associated with lower crash probabilities, severe crash probabilities, and standardized expected-risk estimates for severe and non-severe crashes, with larger reductions observed under AR-HUD. Exploratory subgroup analyses further indicated that driver groups defined by gender, age, and driving experience may exhibit different crash risk characteristics across warning conditions, while AR-HUD generally corresponded to lower crash risk levels across the subgroups. These findings provide a reference for the safety evaluation and further optimization of HUD and AR-HUD warning systems in connected environments. Because the findings were derived from surrogate safety indicators and statistical models based on driving simulation, their applicability to real-world traffic conditions requires further validation using real-vehicle experiments and observed crash data.