深入研究人类驾驶员的驾驶行为和习性，对于推进智能汽车的拟人化决策规划，改善驾驶安全性具有重要意义。针对高速公路这一典型场景，基于NGSIM（Next Generation Simulation）数据集提取有效表征换道驾驶行为的特征参数，分析换道驾驶行为与驾驶参数的相关性，量化驾驶行为特性，建立了基于高斯混合-隐马尔科夫理论（Gaussian mixed model-hidden Markov model，GMM-HMM）的换道意图识别模型。研究结果表明：该模型识别准确率较高，在换道点1.0 s之前的换道行为识别准确率达到95.6%，在有换道意图的时刻识别准确率超过80%，可应用于智能汽车换道策略的拟人化设计，有效降低换道风险，改善驾驶安全。

Understanding human driving behaviors has significant implications for promoting decision-making in intelligent vehicles and improving driving safety. This study focuses on highway lane-changing behavior, using the NGSIM (Next Generation Simulation) Dataset to extract key parameters and analyze the correlation between these parameters and driving behaviors. A GMM-HMM-based model for lane-changing intention recognition was developed, achieving an accuracy of 95.6% in predicting lane changes 1.0 s before they occur, and an accuracy of over 80% in recognizing lane-changing intentions. This model can be applied to intelligent vehicle design to effectively reduce lane-changing risks and improve driving safety.

U448.213

国家自然科学基金资助项目（51875061）。