Identifying and Extracting Pedestrian Behavior in Critical Traffic Situations

TL;DR

The paper tackles identifying critical pedestrian–vehicle interactions in real-world traffic by extracting them from camera-based observations. It introduces a two-tier approach that combines a space-sharing conflict metric (PET) with a novel pedestrian motion adaption metric to filter a large baseline of interactions down to a small, high-quality set of critical cases. From 110 hours of traffic video, they reconstruct 11,089 pedestrian trajectories and derive a baseline catalog of 20,978 potential conflicts, narrowing to 21 candidates and 7 confirmed critical interactions after motion-adaption analysis; the critical cases are publicly available to support pedestrian-model development. The results demonstrate that late-stage pedestrian responses to approaching vehicles can be effectively identified in real-world data, providing valuable datasets and methods for calibrating scenario-based safetyModels and systems.

Abstract

A better understanding of interactive pedestrian behavior in critical traffic situations is essential for the development of enhanced pedestrian safety systems. Real-world traffic observations play a decisive role in this, since they represent behavior in an unbiased way. In this work, we present an approach of how a subset of very considerable pedestrian-vehicle interactions can be derived from a camera-based observation system. For this purpose, we have examined road user trajectories automatically for establishing temporal and spatial relationships, using 110h hours of video recordings. In order to identify critical interactions, our approach combines the metric post-encroachment time with a newly introduced motion adaption metric. From more than 11,000 reconstructed pedestrian trajectories, 259 potential scenarios remained, using a post-encroachment time threshold of 2s. However, in 95% of cases, no adaptation of the pedestrian behavior was observed due to avoiding criticality. Applying the proposed motion adaption metric, only 21 critical scenarios remained. Manual investigations revealed that critical pedestrian vehicle interactions were present in 7 of those. They were further analyzed and made publicly available for developing pedestrian behavior models3. The results indicate that critical interactions in which the pedestrian perceives and reacts to the vehicle at a relatively late stage can be extracted using the proposed method.

Figures (8)

• Figure 1: Example of an interactive pedestrian behavior, which is denoted by the speed profile over time. The moment in which the pedestrian perceives the vehicle is denoted as $t_p$, which is followed by a delayed reaction at moment $t_d$, where the pedestrian starts to decelerate. The $t_f$ indicates the moment when the pedestrian's deceleration is finished. From the vehicle perspective, the decreases in the course of the interaction.
• Figure 2: Orthophoto of the selected observation point St. Peter Schulzentrum. The observed traffic site is a signal-controlled pedestrian crosswalk, where pedestrians frequently rush to catch buses and trams.
• Figure 3: Aligned 3D scene representation of the traffic site. The areas in red denote the pedestrian zones (which can either be approach and target), the green area represents the for .
• Figure 4: The red boxes represent the pedestrian $P_i$, the blue boxes the vehicle $V_j$. The yellow area, in which the two contours overlap is the . For this specific conflict between $P_i$ and $V_j$, the pedestrian is moving from zone 2 to 1, and the vehicle is approaching from the far-side lane. The is negative as the vehicle approaches first. For the vehicle $V_k$ the pedestrian is in the before the vehicle approaching from the near-side lane.
• Figure 5: Exemplary curve fitting results in the least square sense, on two different pedestrian speed profiles of the reconstructed trajectories. The solid red line represents a pedestrian who is adapting the speed profile used, while the orange line presents an ordinary walking behavior. The $\operatorname{std}(\mathbf{e})$ is 0.049m/s for the non-ordinary case and 0.0067m/s for the other. The considered time period $T_{ROI}$ is indicated by the fully opaque lines, the faded parts represent the speed profile within the approach and target zone. The approximation $\hat{v}$ is outlined as dashed line.