驾驶员无意识车道偏离识别方法

Recognition method of driver′s unintentional lane departure

为了改进车道偏离预警系统的工作效能,本文提出了考虑人-车-路特性的无意识车道偏离识别方法。首先,明确了无意识车道偏离识别的具体含义,将其划分为疲劳车道偏离和次任务车道偏离;其次,利用受试者工作特性曲线(ROC)确定无意识车道偏离的识别时间窗口,保证了无意识偏离样本筛选的有效性;再次,以12名驾驶人为试验对象,采集并对比分析了驾驶员操纵特性、车辆运动状态和车辆与车道线相对运动状态等相关参数,并分别选取作为疲劳车道偏离和次任务车道偏离识别基本特征;最后,采用高斯混合隐马尔科夫模型(GM-HMM)构建无意识车道偏离识别模型。实验结果表明,本文方法具有较好的识别效果。

In order to improve the performance of lane departure warning system, an unintentional lane departure analysis method is proposed. This method combines the driver＇s operation characteristics, vehicle’s motion characteristics and the relationship between the vehicle and the lane. First, the unintentional lane departure is classified into two parts： lane departure by fatigue and lane departure by secondary task. Then, experiments of unintentional lane departure are carried out through the co-simulation platform based on CarSim and LabWIEW. Twelve drivers of different genders, proficiencies and driving behaviors are selected to participate the experiments. The unintentional lane departure parameters are collected and analyzed, including the driver＇s operation behavior, the motion characteristics of the vehicles and the relative motion position between the vehicle and the lane. Finally, an unintentional lane departure recognition model is constructed based on Gaussian Mixture- Hidden Markov Model （GM-HMM）. The recognition results show good performance of the proposed model in online and offline tests.