基于多模态生理信号的驾驶人脑力负荷评估

深圳大学人因工程研究所,广东深圳518060

交通信息工程及控制;脑力负荷;机器学习;多模态;脑电信号;心电信号;皮电信号

Evaluation of drivers'mental workload based on multi-modal physiological signals
ZHANG Qiliang,YANG Kunhua,QU Xingda,and TAO Da

Institute of Human Factors and Ergonomics, Shenzhen University, Shenzhen 518060, Guangdong Province, P. R. China

traffic information engineering and control; mental workload; machine learning; multi-modal;electroencephalographic signal; electrocardiographic signal; electrodermal activity signal

DOI: 10.3724/SP.J.1249.2022.03278

备注

准确评估驾驶人脑力负荷状态对降低因驾驶人脑力负荷过载导致的交通事故具有重要意义.基于典型驾驶场景,结合N-back认知负荷次任务,设计不同难度的驾驶任务实验,研究驾驶人脑力负荷.实验收集驾驶人在任务完成过程中的多种模态生理信号(脑电、心电和皮电信号)及美国航空航天局任务负荷指数量表主观脑力负荷数据,提出基于多模态生理信号特征分析和模式识别的驾驶人脑力负荷分类模型,并比较不同模态生理信号及其组合在3种典型机器学习算法(随机森林、决策树和k最近邻模型)中的脑力负荷分类识别效果.研究表明,基于不同模态生理信号组合的脑力负荷分类模型具有不同的分类准确率.单一模态生理信号的分类模型中,基于皮电、心电和脑电信号的分类模型准确率依次增加;基于多模态生理信号的分类模型准确率普遍优于单一模态分类模型;基于脑电、心电及皮电3模态生理信号的随机森林分类模型具有最高的分类准确率.
Accurately assessing the driver's mental workload is of great significance to reduce the traffic accidents caused by the driver's mental overload. This study aims to evaluate drivers' mental workload in simulated typical driving scenarios, with N-back cognitive tasks used to manipulate varied levels of task difficulty. We collect data on multi-modal physiological signals including electroencephalogram (EEG), electrocardiogram (ECG), and electrodermal activity (EDA) signals, and subjective mental load of the National Aeronautics and Space Administration task load index (NASA_TLX) during the task completion process of the driver in the experiment, and propose a series of mental workload classification models based on feature analysis and pattern recognition of the multi-modal physiological signals. These classification models are verified by machine learning algorithms of random forest, decision tree and k-nearest neighbor models. The results show that the accuracy of classification models varies with different modalities of physiological signals. EEG-based classification models yield the highest accuracy among single-modal classification models, followed by EDA-based and ECG-based models. Multi-modal-based classification models generally perform better than single-modal classification models. The random forest classification algorithm based on three-modal physiological signals of EEG, ECG and EDA has the highest classification accuracy.
·