Research on driver fatigue detection via feature fusion

Existing train driver fatigue detection methods usually measure drivers’ biological signals. However, these methods usually use a single biosignal, and don’t consider potential connections between the biosignals. In addition, most biosignal detection methods focus on extracting biosignal features, while ignoring the relationship between the signal itself and fatigue. To address these issues, we integrate the photoplethysmographic and electrodermal activity signals to improve the accuracy of driver fatigue detection. Specifically, we utilize a convolutional neural network to fuse the two signals at the data layer to achieve multimodal fusion fatigue detection. The results show that the two signals intercepted for a 5-minute time window are fused in parallel at the data layer as inputs to the CNN model, the accuracy reaches 95.83%. The results of this study can be applied to the development of reliable fatigue detection methods for train drivers to improve driving traffic safety.