Empirical Evidence of Bounding Box Applications for Improved Surrogate Safety Measures

Abstract: Estimating Surrogate Safety Measures (SSMs) plays a significant role in assessing road safety by identifying potential vehicle conflicts proactively. Traditional approaches to SSM estimation often rely on single-point vehicle coordinates, such as the vehicle's center point or frontal edge. While effective in lane-based homogenous traffic settings, these simplified methods can lead to significant inaccuracies in area-based traffic with diverse traffic composition. The issues include, but are not limited to, inaccurate estimation of relative distances between vehicles, underestimating critical interactions, and failing to account for vehicle size and shape variations. Such shortcomings can distort the true safety depiction, particularly when applied to traffic environments with diverse vehicle types and area-based traffic flows.
To address the highlighted limitations, this study proposes a generalized methodology that incorporates bounding boxes (BB) to capture the varied vehicle geometries and spatial movements of vehicles more accurately. Rather than relying on a single point, the bounding box surrounds the vehicle, enabling a more precise estimation of distances and interactions between vehicles. This approach is versatile and can be applied across various traffic conditions, including lane-based and as well as area-based traffic settings. By accounting for the actual geometries of vehicles, the BB approach rectifies several issues with existing SSM estimation techniques, such as misidentifying conflicts and over-filtering critical vehicle interactions by a fixed threshold.
The methodology was empirically validated using high-fidelity vehicle trajectory data collected via drones, which captured traffic on a six-lane divided urban arterial. Two different methods were applied to corroborate the methodology with empirical data: the traditional point-based approach and the proposed bounding box method. The results demonstrate that the bounding box approach leads to more accurate identification of safety-critical interactions. SSMs such as Time to Collision (TTC) and Proportion of Stopping Distance (PSD) were used to compare the performance of both methods. The BB approach showed a marked improvement in detecting critical conflicts and provided a more reliable representation of traffic safety conditions.
This research highlights the existing fallacies in practice and the importance of incorporating more detailed vehicle representations in SSM estimation, offering a robust framework that is adaptable to various traffic environments. By refining the tools available for proactive safety assessments, this methodology enables better identification of potential risks, leading to more effective interventions to reduce traffic-related accidents. The proposed approach has broad applicability and can enhance road safety in complex traffic environments.