Research and Implementation of a Fatigue Driving Detection System Based on Visual Features

Abstract

With the global surge in car ownership, fatigue driving has become a leading cause of traffic accidents, posing severe threats to road safety and creating an urgent need for high-precision, real-time detection solutions. To address the limitations of existing technologies such as single-feature reliance, poor environmental adaptability, and inadequate on-board deployment capability, this study designs and implements a fatigue driving detection system based on multi-dimensional visual feature fusion. The system integrates image processing, pattern recognition, and machine learning technologies: it adopts Contrast-Limited Adaptive Histogram Equalization (CLAHE) and hybrid median-Gaussian filtering for image preprocessing to enhance quality under complex lighting and weather conditions; utilizes Multi-task Cascaded Convolutional Networks (MTCNN) for eye region localization and PERCLOS index calculation, PoseNet for 3D head pose estimation, and Dlib for facial key point extraction to capture comprehensive fatigue-related features; and employs a Back Propagation (BP) neural network for multi-feature fusion, optimized through model pruning and quantization to achieve lightweight deployment. Experimental results demonstrate that the system achieves 92.3% detection accuracy, a false alarm rate of 3.7%, and an average processing delay of 32.6ms (maximum <40ms) on the Jetson Nano embedded platform, with an inference speed of 25 FPS. It maintains strong robustness under complex environments (e.g., strong light, night, rainy and foggy days) with an accuracy fluctuation of less than 5%. This study provides a reliable, non-invasive technical solution for traffic safety, which is of great significance for the development of intelligent transportation systems and the reduction of fatigue-related accidents.