Rear-end crash prevention depends strongly on warning timing, yet most connected warning studies still rely on fixed trigger rules that are only loosely aligned with drivers' perceptual urgency and are rarely evaluated across different car-following conditions under degraded visibility. This study evaluated a vehicle-to-vehicle-enabled connected collision warning (CCW) strategy triggered by the visual-looming-based tau-1 metric, which provides a perception-consistent indicator of time-critical risk. A driving simulator experiment was conducted under rainy conditions, in which 37 licensed drivers were exposed to 12 lead vehicle deceleration events each. Behavioral responses and surrogate safety outcomes were analyzed using linear mixed-effects and mediation models. Relative to the no-warning condition, the tau-1-triggered CCW reduced collision occurrence and shortened brake reaction time by 10.1% at far headway and 7.5% at near headway, while producing limited changes in brake transition time, maximum deceleration, and the proportion of uncomfortable deceleration. The safety effects of the warning were jointly moderated by headway and lead vehicle deceleration, with stronger benefits observed under moderate urgency and attenuated effects under extreme urgency. Mediation analysis further indicated that 39% of the reduction in collision severity was transmitted through brake reaction time, whereas 61% was attributable to direct warning effects on post-braking control. These findings support perception-consistent and scenario-adaptive trigger calibration for connected warning systems, in which warning thresholds should be adjusted according to headway condition, lead vehicle deceleration intensity, and degraded-visibility risk rather than relying on a single fixed time-to-collision (TTC) or distance threshold.
Addressing macroscopic road safety complexities requires understanding regional factor interactions. To bridge static performance benchmarking and dynamic causation analysis, this study proposes a two-stage "Performance Evaluation-Driving Factor Diagnosis" framework. Utilizing 2023-2024 panel data exclusively from the National Bureau of Statistics for 30 Chinese provinces, safety indicators are decoupled into an Evaluation Level and a Diagnostic Level. Performance is objectively stratified using a multicriteria pipeline (MEREC, PROMETHEE II, and K-Means). Subsequently, Apriori association rule mining extracts multidimensional associative configurations from both static and dynamic perspectives. Results reveal significant systemic inertia, with 76.67% of provinces maintaining stable safety ranks. Static mining demonstrates explicit functional pathways: physical infrastructure primarily suppresses accident frequency, whereas robust medical resources serve as a critical containment mechanism against severe casualties. Dynamically, upward rank mobility requires the synergistic expansion of both road infrastructure and medical capacity. Conversely, rank declines are precipitated by resource-risk mismatches, such as medical personnel deficits relative to escalating traffic exposure. Ultimately, this research provides an evidence-based approach for precision safety governance, transitioning from generic interventions to targeted capacity building tailored to regional structural characteristics.
Surrogate safety analysis turns trajectories into events and models the high-risk ones. What counts as an event is rarely reported as a modeling decision, yet it shapes the population and every downstream inference. Using 27,219 lane-change markers from the exiD drone dataset, we apply three semantic filters (lenient, baseline, strict) that reduce the modeling-ready set from 12,925 events to 2,609, an 80% drop. A composite surrogate-risk label built from seven indicators supports two models: a full model with access to those indicators, and a reduced model without them, evaluated under location-transfer validation with calibration. The full model reconstructs the label (AUC 0.980-0.989) under both definitions; the reduced model drops from AUC 0.709 to 0.630, with a three-way control assigning about 42% of the drop to sample size and class imbalance, about 2% to within-set relabeling, and 56% to the analyzed-population change. The direction holds in all twelve cells of a grid of alternative strict-filter thresholds. Calibration improves probability reliability but does not recover the lost discrimination: under the strict definition fixed-threshold classification collapses while ranking weakens. Event construction is part of the measurement model, and reporting accuracy without event-definition sensitivity overstates what trajectory-based models can claim.
Dangerous lane changes in urban expressway merging areas involve multi-vehicle interactions and coupled lateral-longitudinal risk, which instantaneous longitudinal surrogate safety measures alone cannot capture. This study proposes a framework jointly capturing longitudinal car-following risk and lateral intrusion risk. Vehicle trajectories were extracted from UAV videos at an urban expressway merging area, yielding 206 lane-change samples after screening and reconstruction. Longitudinal safety distance indices were built using a stopping sight distance model, and lateral indices via a responsibility-sensitive safety model. Risk exposure and severity were calculated per dimension and integrated through a fault-tree structure into a composite risk index. The classification threshold was determined using a probability-weighted moments method, with robustness evaluated through bootstrap resampling. The framework identified 53 dangerous lane changes (25.7% of samples). Cross-scenario validation using near-crash events from the 100-Car Naturalistic Driving Study achieved 100% recall, versus 69.2% for a conventional TTC-based method. Risk decomposition showed that TTC-missed events were predominantly lateral-driven, demonstrating superior capability in detecting longitudinally safe but laterally dangerous scenarios. These findings provide a quantitative foundation for identifying dangerous lane changes and supporting proactive safety management in expressway merging areas.
Identifying aggressive longitudinal driving behaviors among heavy goods vehicle (HGV) drivers is important for proactive freight safety management. This study proposes a context-aware framework for identifying aggressive acceleration and braking behaviors using naturalistic GPS trajectory data from 97 HGVs. Acceleration and braking events were extracted from continuous trajectory records and characterized using multidimensional kinematic features, including velocity, acceleration/deceleration, jerk, variability, and duration. Principal component analysis was first applied to group the extracted features into velocity-related and maneuver-related dimensions. A two-stage K-means clustering framework was then developed, in which driving events were first divided into slow- and fast-velocity environments and subsequently classified into gentle, moderate, and aggressive driving styles within each velocity condition.The results were validated using clustering indices, statistical tests, external separation, FCW-based safety correlation, and microscopic behavior representation. Aggressive acceleration accounted for 7.7% and 3.4% of slow- and fast-velocity acceleration events, respectively, whereas aggressive braking accounted for 9.7% and 7.0% of braking events. Slow-velocity aggressive behaviors showed higher maneuver intensity and variability, and aggressive braking was more closely associated with FCW warnings than aggressive acceleration. Microscopic behavior representation further revealed heterogeneous individual patterns. These findings support driver risk assessment, behavior-specific feedback, and proactive fleet safety management.
