Fatigue Driving Identification and Blood Oxygen Pulse Detection

Abstract

Driver fatigue is a major factor in road traffic accidents. It reduces attention, perception, and reaction time. Existing detection methods- vision-based, behavior-based, or using physiological signals- face issues like environmental sensitivity, individual variability, and limited robustness.  Single-modality approaches are unreliable under diverse driving conditions. To address these challenges, this study proposes a multimodal deep learning framework. It integrates vision-based features with photoplethysmography (PPG)-based signals for robust detection. The system uses a convolutional neural network (CNN) to extract features from facial images. A recurrent neural network (RNN) captures temporal dynamics from physiological signals. Feature-level and decision-level fusion strategies are combined to boost detection accuracy and robustness. The system  monitors multiple fatigue-related indicators at the same time. These include eye aspect ratio, blink frequency, head pose, yawning behavior, heart rate, heart rate variability, and blood oxygen saturation. Experimental evaluations were conducted in both simulated driving environments and real-world road conditions with 30 participants. The results show that the proposed method achieves 92.3% accuracy in simulation and 85.7% in real-world scenarios. This outperforms single-modality approaches by 8-12%. The system also maintains a low latency of less than 150 ms, which meets real-time application requirements. The proposed framework offers a non-intrusive, robust, and scalable solution for driver fatigue detection. It can be effectively integrated into intelligent transportation systems, Internet of Vehicles (IoV), and Advanced Driver Assistance Systems (ADAS). This work help improves  road safety and advances multimodal AI applications in smart mobility.