From Real-World Traffic Data to Relevant Critical Scenarios

TL;DR

The paper presents a data-driven pipeline to identify safety-critical highway lane-change scenarios for ADAS/AD validation by combining real-world in-car and aerial data, defining safety-criticality via observable driving dynamics and established metrics, and applying these metrics to characterize lane-change maneuvers. It then demonstrates synthetic scenario generation through sampling in a traffic simulator (Vissim) and an MIS concept evaluated within the OCTAS framework, illustrating how to extend limited real-world data with data-driven, safety-relevant variants. The work emphasizes cross-source data harmonization, robust lane-change detection, and the use of criticality metrics to prioritize validation efforts, with practical implications for highway automation and safety assurance. Overall, it provides a concrete methodology for building and enriching a safety-critical scenario database and for testing reactive safety mechanisms in simulated environments.

Abstract

The reliable operation of autonomous vehicles, automated driving functions, and advanced driver assistance systems across a wide range of relevant scenarios is critical for their development and deployment. Identifying a near-complete set of relevant driving scenarios for such functionalities is challenging due to numerous degrees of freedom involved, each affecting the outcomes of the driving scenario differently. Moreover, with increasing technical complexity of new functionalities, the number of potentially relevant, particularly "unknown unsafe" scenarios is increasing. To enhance validation efficiency, it is essential to identify relevant scenarios in advance, starting with simpler domains like highways before moving to more complex environments such as urban traffic. To address this, this paper focuses on analyzing lane change scenarios in highway traffic, which involve multiple degrees of freedom and present numerous safetyrelevant scenarios. We describe the process of data acquisition and processing of real-world data from public highway traffic, followed by the application of criticality measures on trajectory data to evaluate scenarios, as conducted within the AVEAS project (www.aveas.org). By linking the calculated measures to specific lane change driving scenarios and the conditions under which the data was collected, we facilitate the identification of safetyrelevant driving scenarios for various applications. Further, to tackle the extensive range of "unknown unsafe" scenarios, we propose a way to generate relevant scenarios by creating synthetic scenarios based on recorded ones. Consequently, we demonstrate and evaluate a processing chain that enables the identification of safety-relevant scenarios, the development of data-driven methods for extracting these scenarios, and the generation of synthetic critical scenarios via sampling on highways.

Figures (8)

• Figure 1: Aerial image patch with detected cars, trucks, and annotated lanes. It can be observed that in spite of the highly accurate geo-referencing, the bounding box positions of large vehicles exhibit a lateral bias in the lane based on perspective effects.
• Figure 2: Object detection in aerial images through oriented bounding boxes can lead to unilateral overestimation of object size in range direction through perspective effects, and thus imply erroneous overlaps with other lanes (shown here as an extreme example at steep off-nadir angles $\nu$ and low altitudes $H$ during final approach). Effects depend on off-nadir distances, object heights $h$, but also object vertical shape. While bounding box precision is high ($\approx$ 0.2 m), the constant per-object bias can be considerably larger ($\approx$ 1 m under described regular acquisition conditions at cruising altitudes, cf. eisemann2023approach).
• Figure 3: Comparison of the number of detected lane changes using the distance criterion (dotted lines) and the peak criterion (dashed lines). For comparison, the actual number of executed lane changes (ground truth in solid lines) is presented. The comparison was conducted for three drivers ($D_1$ in purple, $D_2$ in blue, $D_3$ in green) of the JUPITER vehicle. The number of detected lane changes was examined depending on artificially imposed inaccuracies on the measurement results in the form of Brownian noise (top) and translations (bottom) regarding the vehicle's position determination.
• Figure 4: Lane change durations (top) and speed at the peak point of the lane crossing (bottom) for lane changes of cars (red) and trucks (blue) observed in in-vehicle (left) and bird's eye view (right) data. Median values of all identified lane changes for each driver or recording (horizontal lines) are depicted within the left side of each figure, and the summary of all recordings on the rightmost entry (separated by a vertical solid line) with 25 and 75 percentiles (blue and red boxes), maximum range (whiskers), and outliers (black circles). Dates for the aerial recording are given on the x-axis in the year/month/day format.
• Figure 5: Most critical metric values of lane changes based on the evaluation of criticality metrics for the aerial recordings. Minimum metric values were calculated for $d$, $\mathit{THW}$, and $\mathit{DCE}$ and maximum metric values for $v$, $a_{\textrm{\upshape{lon}}}$, and $a_{\textrm{\upshape{lat}}}$. $\mathit{DCE}$ values were only computed when $\mathit{TTCE} < 2.6$ s. Criticality thresholds are shown by dashed vertical lines. Non-critical areas are shaded in gray. Black arrows indicate the direction towards higher critical values.