Hybrid GNN-Based Driver Monitoring System using Facial Landmarks

M Poojitha1, M Shobha Rani2, K Saketh Ram3 , A Bhagyaraja4, Mrs L Lavanya 5
1 Department of Computer Science and Engineering(AI&ML), Student, Sri Venkateswara College of
Engineering, Tirupati
2Department of Computer Science and Engineering(AI&ML), Student, Sri Venkateswara College of
Engineering, Tirupati
3Department of Computer Science and Engineering(AI&ML), Student, Sri Venkateswara College of
Engineering, Tirupati
4Department of Computer Science and Engineering(AI&ML), Student, Sri Venkateswara College of
Engineering, Tirupati
5Department of Computer Science and Engineering(AI&ML), Assistant Professor, Sri Venkateswara College of
Engineering, Tirupati

Abstract---Driver fatigue and stress are major causes of road accidents and unsafe driving conditions. Traditional methods that rely on Eye Aspect Ratio (EAR) and Mouth Aspect Ratio (MAR) often fail under real-world conditions such as lighting variations, head movements, and subtle facial changes. This paper presents a hybrid driver monitoring framework based on insights from existing studies for detecting both driver drowsiness and stress using facial landmarks. The system utilizes MediaPipe Face Mesh to extract 468 facial landmarks and computes classical features such as EAR and MAR for fatigue detection. In addition, a Graph Neural Network (GNN) is employed to model the relationships between facial landmarks and capture complex behavioral patterns. The outputs of classical and graph-based methods are fused to determine the driver’s state as SAFE, WARNING, or CRITICAL. The proposed hybrid approach aims to balance computational efficiency and intelligent feature learning, making it suitable for realtime applications.