Fingerprint of a Traffic Scene: an Approach for a Generic and Independent Scene Assessment

TL;DR

This paper introduces a universal, scene-independent evaluation model and new metrics for assessing traffic scenes in automated vehicle safety testing, validated with real motion dataset data.

Contribution

It proposes a novel multidimensional evaluation model and two enhanced universal metrics for traffic scene assessment, independent of scene type.

Findings

The evaluation model effectively assesses different traffic scenes.

The new metrics demonstrate universality and robustness.

Validation with real data confirms practical applicability.

Abstract

A major challenge in the safety assessment of automated vehicles is to ensure that risk for all traffic participants is as low as possible. A concept that is becoming increasingly popular for testing in automated driving is scenario-based testing. It is founded on the assumption that most time on the road can be seen as uncritical and in mainly critical situations contribute to the safety case. Metrics describing the criticality are necessary to automatically identify the critical situations and scenarios from measurement data. However, established metrics lack universality or a concept for metric combination. In this work, we present a multidimensional evaluation model that, based on conventional metrics, can evaluate scenes independently of the scene type. Furthermore, we present two new, further enhanced evaluation approaches, which can additionally serve as universal metrics. The metrics we introduce are then evaluated and discussed using real data from a motion dataset.