Abstract:
The traffic accidents are mainly related to unfavorable driving status, e.g. fatigue, stress, distraction and sleepiness, which may result in a considerable amount of vehicle collisions and casualties every year. Generally, stress may be taken as a normal response of human body to dangerous or difficult events. With the development of wearable sensor and wireless technique, researchers are paying more attention to the physiological measures that are highly correlated to driver’s mental states. In-vehicle intelligent systems, the physiological measures can be utilized automatically in various manners to help drivers better manage their negative driving status. However, the driving stress detection based on multimodal physiological signals might affect the driving comfort of drivers, and traditional physiological feature extraction techniques largely rely on the prior knowledge. Aiming at the above shortcoming, this paper proposes a new driving stress detection method by means of single module physiological signal, i.e., GSR signal on the foot (FGSR). Firstly, the abstract features are generated by unsupervised feature learning using 1D-convolutional auto-encoder (CAE), and are further sent to four different base classifiers, i.e., k-nearest neighbor (KNN), gradient boosting decision tree (GBDT), support vector machine (SVM) and random forest (RF). And then, the outputs of different base classifiers are integrated via the voting method to improve the stability and accuracy of driving stress detection. Finally, the proposed model is validated via the MIT-drivedb data set, from which the features extracted from GSR using convolutional auto-encoder have good representational ability for the driving stress. Moreover, the ensemble of different base classifiers can effectively improve the accuracy of final recognition results.