Unsafe lane change behaviors have negative impacts on traffic safety. Identifying these risky patterns can help drivers make safe lane change decisions. In this paper, we develop a two-dimensional indicator based on field theory to evaluate lane change risks on the highway. The driving risk field is constructed by integrating the individual and interaction features of the source vehicle and the interacting vehicle. Parameter calibration with highD trajectory data allows the risk field forces to objectively reflect the risk perception. An adaptive optimization algorithm with cumulative gradient bias correction is introduced for parameter calibration. The enhanced first-order and second-order momentum, along with the long-term memory, are developed to adaptively adjust the learning rate and prune abnormal gradients. A Gaussian mixture model is applied to cluster the calibrated two-dimensional risk indicators, generating the F-F diagram. Experimental results demonstrate the superiority of the proposed driving risk field model and the gradient-based algorithm. Confidence region estimation of the lane change risk distribution further validate the objectivity and interpretability of the F-F diagram. The 95% confidence level offers valuable insights for proactive traffic safety management, and lays a foundation for developing personalized lane change strategies. In the future, the F-F diagram could be integrated into lane change advisory systems or advanced driving assistance systems, helping drivers make informed lane change decisions.
F-F Diagram: A Two-Dimensional Surrogate Safety Indicator Based on Field Theory for Lane Change Risk Assessment
IEEE Transactions on Intelligent Transportation Systems ; 26 , 7 ; 10694-10709
2025-07-01
3556806 byte
Article (Journal)
Electronic Resource
English
Lane change risk assessment and personalized lane change decision method
European Patent Office | 2023
|