中国科学院心理研究所机构知识库

为准确识别驾驶人对潜在危险事件的心理预期,提出一种通过脑电信号对驾驶人心理预期进行识别的方法。参照心理学预期行为研究常用的标准S1-S2范式,改进设计了路侧停靠公交车造成视觉遮挡的人车碰撞事故模拟驾驶试验。模拟任务中以路侧停靠的公交车为线索刺激,公交车头的行人为目标刺激,诱发驾驶人的心理预期。为有效识别驾驶人心理预期,首先采用释放油门的避险行为对每个试次标定是否产生心理预期,然后通过快速傅里叶变换提取相关的脑电特征数据,并通过差异性分析及主成分分析算法对脑电特征指标进行筛选和压缩,最后基于支持向量机建立驾驶人心理预期识别模型。研究获取36名驾驶人数据,共计1 440个样本。结果表明:当驾驶人产生心理预期活动时,枕区和额区的α波能量值显著降低,而β波能量值则显著增加;差异性分析显示共有31项脑电指标对驾驶人心理预期敏感,所有脑电特征指标通过PCA算法进行降维,抽取出5个主成分作为识别模型输入;选择径向基核函数构建SVM识别模型,通过粒子群寻优算法对模型进行优化,模型对驾驶人心理预期水平的平均识别正确率为82.02%,平均AUC面积为0.82,结果表明模型具有良好的识别能力和稳定性,可为驾驶辅助系统的研发提供技术理论支撑。

In order to accurately identify the psychological anticipation of drivers to potentially hazardous events，a classification model based on electroencephalogram(EEU) signal as tnputs is proposed in the present study. A simulation driving experiment was modified by referring to a standard psychological S1-S2 paradigm，which is widely used for studying anticipatory behaviors. There were some potential car-pedestrian crush events because of the parked bus beside the road during the experiment. In this simulation，the parked bus was the cue stimuli and the pedestrian was the target stimuli. This setting intended to arouse the driver's psychological anticipation. Firstly，in order to recognize driver's anticipatory activities, the throttle release action in the behavior data was utilized to label the trial. Later，the related EEU feature data was extracted by Fast Fourier Transform(FFT), and the FFG feature indexes were selected and compressed through analysis of variance and principal component analysis(PCA)algorithms，respectively. Finally，based on support vector machine (SVM)，a driver's psychological anticipation state classification model was established. Data for 36 drivers were collected in the experiment，and a total of 1 440 samples were collected. The results show that the energy of a band in the occipital and frontal regions decreases significantly when the drivers arouse their anticipatory activities whereas the energy of β band increases significantly. 31 EEU features are sensitive to the driver's anticipatory activities，and 5 principal components were extracted as inputs of classification model by PCA. An SVM model was constructed with RBF kernel function，and the parameters were optimized using the particle swarm optimization algorithm. By using EEU features as inputs to identify the psychological anticipation state of the driver，it was observed that the average correct rate is 82. 02% and the average area under curve(AUC) is 0.82. This means that the model is stable and performs with a satisfactory recognition ability; therefore，it can provide technical and theoretical support for the development of driving assistance system.

中国铁路总公司科技计划项目(2018F024);; 四川省科学技术重点研发项目(2019YFG0043)~~