A Remote End-to-end Real-time Heart Rate Detection Method Based on rPPG in Driving Scenarios

Driver heart rate (HR) detection is a significant research topic, and real-time understanding of the driver's physiological state can significantly improve driving safety. Camera-based remote photoplethysmography (rPPG) technology can be an alternative to inconvenient wearable devices for noncontact HR detection. However, although rPPG technology has been extensively researched, its application to driving scenarios requires overcoming various disturbances in outdoor conditions, particularly the effects of light variations. To apply rPPG technology to driving scenarios, we propose two improvements. First, we utilize an exposure correction model to preprocess the input image to minimize the effect of illumination variations on the robustness of the prediction. Second, we propose an end-to-end HR detection network based on the improvement of Rhythmnet, which makes it possible to achieve real-time HR detection without the need to perform additional image processing steps. Compared with existing HR detection methods for driving scenarios, our method achieved significant improvements, and experiments on the MR-NIRP Car public dataset achieved the best or near-best results.