基于生理特征映射的驾驶员情绪在线识别模型构建方法

高精度驾驶员情绪识别模型是智能车辆安全辅助驾驶系统构建的关键问题之一。针对该问题，提出一种基于生理特征映射的驾驶员情绪在线识别模型构建方法。对不同情绪下驾驶员心电、脉搏生理信号进行时域特征提取，建立生理信号-情绪关联样本库，给出改进式自编码(Improved stacked autoencoder, I-SAE)神经网络架构，构建基于I-SAE神经网络的人员情绪生理表征和离线辨识模型；同步采集行驶过程中不同情绪下驾驶员生理信号和车辆状态数据，利用主成分分析法选取车辆状态特征参数向量，通过I-SAE模型识别处理生理信号数据，将驾驶员情绪的生理表征映射为车辆状态表征，构建车辆状态-驾驶员情绪关联样本库；在此基础上，建立基于学习矢量量化(Learning vector quantization, LVQ)的驾驶员情绪在线识别模型。试验数据表明，该方法构建的驾驶员情绪在线识别模型正确识别率达84%以上，可满足安全距离预警等车辆智能系统需要。

Online Recognition Model Construction Method of Driver Emotion Based on Physiological Feature Mapping

High-precision driver’s emotion recognition model is one of the key problems for intelligent vehicle safety assisted driving systems. To solve this problem, a method is proposed to build driver’s emotion online recognition model through physiological feature mapping. The time-domain features of driver’s physiological signals of ECG and pulse under different emotions are extracted, the sample library of physiological signal and emotion correlation is established. An improved stacked autoencoder (I-SAE) neural network is proposed, an emotion model of physiological representation and off-line identification based on I-SAE is built. Both physiological signals and vehicle state data are collected for different emotions during the driving. The vehicle state characteristic parameter vector is made by principal component analysis and the physiological signal data is processed through I-SAE model. The driver’s emotion representation is transferred from physiology to vehicle state, and the correlation sample library of vehicle state and driver emotion are obtained. On this basis, a driver emotion online recognition model based on learning vector quantization (LVQ) is established. The experimental data show that the correct recognition rate of the driver's emotion online recognition model constructed by this method is more than 84%, which can meet the needs of intelligent systems such as vehicle safety distance early warning.