Review of Graph-Based Hazardous Event Detection Methods for Autonomous Driving Systems

Automated and autonomous vehicles are often required to operate in complex road environments with potential hazards that may lead hazardous events causing injury or even death. Therefore, a reliable event detection system is key enabler for highly (e.g., Level 4 5 vehicles) without human supervision significant periods of time. One promising solution the problem use graph-based methods powerful tools relational reasoning. Using graphs organise heterogeneous knowledge about operational environment, link scene entities users, static objects, traffic rules) describe how they affect each other. Due growing interest opportunity presented by detection, this paper provides comprehensive review state-of-the-art we categorise as rule-based, probabilistic, machine learning-driven. Additionally, present an in-depth overview available datasets facilitate training evaluation metrics assess model performance. In doing so, aim provide thorough insight into research opportunities open challenges.