Pedestrian Behavior Prediction for Automated Driving: Requirements, Metrics, and Relevant Features

TL;DR

The requirements on pedestrian behavior prediction for automated driving via a system-level approach is analyzed via a large-scale dataset comprising thousands of real-world pedestrian-vehicle interactions with human drivers and a novel metric tailored to measure prediction performance is proposed.

Abstract

Automated vehicles require a comprehensive understanding of traffic situations to ensure safe and anticipatory driving. In this context, the prediction of pedestrians is particularly challenging as pedestrian behavior can be influenced by multiple factors. In this paper, we thoroughly analyze the requirements on pedestrian behavior prediction for automated driving via a system-level approach. To this end we investigate real-world pedestrian-vehicle interactions with human drivers. Based on human driving behavior we then derive appropriate reaction patterns of an automated vehicle and determine requirements for the prediction of pedestrians. This includes a novel metric tailored to measure prediction performance from a system-level perspective. The proposed metric is evaluated on a large-scale dataset comprising thousands of real-world pedestrian-vehicle interactions. We furthermore conduct an ablation study to evaluate the importance of different contextual cues and compare these results to ones obtained using established performance metrics for pedestrian prediction. Our results highlight the importance of a system-level approach to pedestrian behavior prediction.

Figures (12)

• Figure 1: Exemplary scenario in which the behavior of a pedestrian depends on multiple contextual cues, e.g. the present road infrastructure or interactions with other traffic participants.
• Figure 2: Dataset statistics: (a) distribution of pedestrian tracks with respect to the three courses; (b) locations at which pedestrians crossed the street.
• Figure 3: Example of labeled pedestrian attributes: The compass plots depict head and body orientation, where the filled upper compass indicates that the pedestrian is looking at the ego-vehicle.
• Figure 4: Semantic map shown on top of a road map. Colors denote different semantic classes. $\copyright$https://www.openstreetmap.org/copyright contributors (https://openstreetmap.org, https://opendatacommons.org)
• Figure 5: Two traffic scenarios of a vehicle approaching a pedestrian crossing the roadway. In scenario (a) the driver brakes in order to maintain a Time-to-Collision ($\text{TTC}$) above approximately $2 \, \mathrm{s}$ while the pedestrian is traversing the driving corridor of the vehicle. In scenario (b) the driver maintains a sufficiently large time gap between his vehicle and the pedestrian without having to brake.