Abstract: A lane-changing hazard perception model based on the behavior characteristics of the driver is established, and a new algorithm is proposed with on-line parameter identification and adjustable threshold. By using the fuzzy logic method, the influence of the surrounding vehicles on lane change is determined by the speed correlation degree, the safety factor and the lateral deviation to modify hazard perception model parameters. Then the model parameters are on-line identified by recursive maximum likelihood estimation, and the real-time risk assessment value is obtained. Finally, based on the information entropy, the optimal alarm threshold is searched, and the real-time evaluation value is compared with the alarm threshold to judge the alarm state of the system. Verification results using natural driving behavior data from real-car experiments show that the accuracy of the adaptive warning model is 92.1% and the time to predict the state of danger can be advanced by 0.3-1s , which accords with the psychological expectation and practical operating characteristics of the driver.

Key words: lane change warning, adaptive driver characteristics, fuzzy logic, maximum likelihood estimation, information entropy