Non-Invasive Driver Monitoring: A contactless rPPG Approach using NIR Camera

Driver fatigue and stress are major causes of road accidents. This work presents a contactless method for driver monitoring using remote photoplethysmography (rPPG) with a Near-Infrared (NIR) camera and active NIR LED illumination. Hardware performance was evaluated using a Synthetic Target Setup and Signal Quality Indices (SQIs). A supervised Separable Convolutional Neural Network (Sep-CNN) and Gated Recurrent Unit (GRU) model was trained on NIR facial videos from the MR-NIRP dataset, achieving a 3.8 bpm MAE on indoor data. Future work includes RGB-NIR fusion, attention-based models, and an internal database covering indoor and outdoor conditions for robust, explainable driver monitoring.