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1s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。

In-depth study of human driver's driving behavior and habits has significant implications for promoting the anthropomorphism of decision-making in intelligent vehicles and improving driving safety. For the typical scenario of highways, effective feature parameters that characterize lane-changing driving behavior were extracted based on the NGSIM dataset. The correlation between lane-changing driving behavior and driving parameters was analyzed, and driving behavior characteristics were quantified. A GMM-HMM-based lane-changing intention recognition model was established. The research results show that the model has a high recognition accuracy. The recognition accuracy of lane-changing behavior 1 second before the lane-changing point reaches 95.6%. The accuracy of recognizing lane-changing intentions exceeds 80% when there is the intention to change lanes. The model can be applied to the anthropomorphic design of intelligent vehicle lane-changing strategies, effectively reducing lane-changing risks and improving driving safety.

U448.213

国家自然科学基金资助项目(51875061)