Evasive behavior-based method for threat assessment in different scenarios: A novel framework for intelligent vehicle.

The threat assessment process is a crucial part of intelligent vehicles (IVs) for evaluating the levels of criticality and taking possible measures to avoid the collision, especially for the collision avoidance systems (CAS). In this study, a novel threat assessment framework based on the driver's evasive behavior, namely the CPIC, is proposed, which integrates the crash probability (CP) and inevitable crash (IC) state to be widely used by different CAS in different scenarios. In the first step of the CPIC framework, the detailed evasive driver behavior models (E-DBMs) in the form of probability density functions (PDFs) were introduced to generate more realistic collision-avoidance trajectories. Two techniques for sampling these trajectories, namely the Markov Chain Monte Carlo (MCMC) and adaptive Gaussian mixture framework (GMM) methods, were utilized to ensure the samples were from the area of high probability density in the E-DBMs. The CP value could be derived by considering multiple collision-avoidance trajectories. To confirm the IC state in step 2, the CPIC framework employed the driving limit-based approach for IC checking, which combined the CP value to double-check the unavoidable collision. A total of 82 critical events from the real-world naturalistic driving study, the Strategic Highway Research Program 2 (SHRP2), were extracted to verify the performance of the CPIC framework in different scenarios. Results show that the proposed method clearly revealed the risk levels when two vehicles were approaching, and 80 events were successfully identified as near crashes/crashes. Moreover, the real-time performance of the CPIC framework was also demonstrated. The findings indicate this CPIC framework could be used in practical applications of IVs in different scenarios.