Safety risk assessment of human–vehicle collaboration in advanced driver assistance systems based on AHP-ANN

Abstract

Advanced Driver Assistance Systems (ADAS) play an increasingly important role in intelligent transportation systems. However, under partial automation conditions, the safety of human–vehicle collaboration is jointly influenced by system perception, driver behavior, vehicle execution, and road environment. Traditional Analytic Hierarchy Process (AHP) can effectively determine the relative importance of risk indicators, but it is limited in capturing nonlinear interactions among multiple factors. To address this issue, this paper proposes an AHP-ANN-based safety risk assessment method for ADAS human–vehicle collaboration. First, a hierarchical indicator system including four first-level dimensions and sixteen second-level indicators is established. Then, AHP is used to derive expert prior weights, which are embedded into an Artificial Neural Network (ANN) as weighted input features. Finally, the proposed model is validated using simulation samples generated in CARLA combined with public driving-behavior references. Experimental results show that driver behavior risk has the highest first-level weight, while takeover reaction time, false alarm and missed alarm probability, driver distraction level, and target detection error are the most critical second-level indicators. In addition, the proposed AHP-ANN model achieves 92.4% accuracy and 0.944 AUC, outperforming both the AHP-only and ANN-only baselines. The results indicate that incorporating expert knowledge into data-driven learning improves both the interpretability and predictive performance of ADAS safety risk assessment. This study provides a theoretical and technical reference for human–machine interaction design, takeover strategy optimization, and safety enhancement of ADAS.