A Comprehensive Survey on Dynamic Perception and Quantitative Assessment of Driving Risks in Intelligent Connected Vehicles

With the rapid popularization of intelligent connected vehicles (ICVs), there is an urgent need for systematic research to address their multidimensional safety risks. Current risk research is predominantly technology-driven, with a strong focus on cybersecurity and data security. However, the increasing proportion of functional safety and SOTIF risks observed in real-world accidents highlights imbalances in research priorities and insufficient dynamic adaptability. Existing risk perception techniques primarily rely on dynamic analysis and multi-modal sensor fusion, while quantitative risk assessment focuses on multi-tier testing frameworks and fuzzy mathematical modeling. Risk-informed decision-making strategies depend on lifecycle management and scenario database iteration, but these approaches are constrained by limited real-time performance and cross-domain collaboration challenges. To overcome these limitations, future efforts should prioritize the development of a multidimensional dynamic risk cognition framework, alongside advancements in vehicle-road-cloud collaborative decision-making, lightweight AI models, and standardized ecosystems. These innovations aim to address critical technical bottlenecks such as extreme scenario coverage and cross-domain risk mapping, thereby providing theoretical foundations and industrial implementation pathways for enhancing ICV safety.