基于视觉特性的驾驶人换道意图识别

Drivers' Lane Changing Intention Identification Based on Visual Characteristics

为了将汽车换道辅助系统的预警时刻提前,改进辅助系统的工作效能,提出了利用驾驶人不同任务下眼动行为差异性识别换道意图的方法。在实际道路环境中进行试验,记录16名驾驶人共计408次换道意图阶段和403次车道保持阶段的眼动行为数据,对比了2个阶段中驾驶人眼动行为数据的差异,构建了包含7个参数的换道意图识别指标体系。运用证据理论确定识别框架和证据链,构造基于广义汉明距离的基本信任分配函数,建立了基于多证据融合识别的换道意图识别模型,并分析了识别信度和时序性。结果表明:该模型识别结果的灵敏度、特异度和准确率分别为89.86%,90.21%和90.02%,具有较好的识别效果。

In order to warn the drivers earlier and improve the performance of lane changing assistance system,using the difference of drivers＇eye movement behaviors under different tasks,a new method to identify drivers＇ lane changing intention was proposed.In actual road environment,16 drivers＇ eye movement data were collected,in which there were more than 408 samples of lane changing intention stages and 403 samples of lane keeping stages.The lane changing intention identification index system including seven parameters was established by analyzing the difference of drivers＇ eye movements in the two stages.With introducing the evidence theory,the identification frame and evidence chain were determined.The basic trust distribution function based on the generalized Hamming distance,and lane changing intention identification model based on multi-evidence fusion identification was established,and the identification reliability and sequence were analyzed.The sensitivity,specificity and accuracy is 89.86%,90.21% and 90.02%,respectively,which shows a good identification effect.