Improving automatic detection of driver fatigue and distraction using machine learning

Wu, Dongjiang · 2024 · arXiv

archive: archived pipeline: cataloged verified

Get this paper ↗ (full text — opens at the source; we link to it, we don't host it)

Summary

This study addresses the critical safety issue of driver fatigue and distracted driving, which are leading causes of traffic accidents. Motivated by the limitations of physiological and vehicle-data-based detection methods, the author proposes a vision-based approach using machine learning to simultaneously detect these behaviors. The research aims to improve upon existing methods by developing a system that offers better accuracy and computational efficiency, suitable for integration into Advanced Driver Assistance Systems (ADAS). The methodology employs two distinct computer vision techniques. For fatigue detection, the system utilizes a Facial Alignment Network (FAN) to identify 68 facial feature points, calculating the distance between points to monitor eye and mouth opening/closing. This data is used to compute the PERCLOS (percentage of eye closure) metric, a standard indicator of drowsiness. For distraction detection, the study implements a Convolutional Neural Network (CNN) based on the MobileNet architecture, chosen for its lightweight design and suitability for real-time processing on resource-constrained devices. The model was trained and tested using a combination of public datasets, specifically the State Farm Distracted Driver Detection (SFDDD) dataset, and a custom five-class dataset collected via a front-facing webcam. Data preprocessing included cleansing, denoising, calibration, and alignment to ensure consistency. The results demonstrate that the proposed system effectively identifies various distracted driving behaviors and fatigue states. The MobileNet-based CNN achieved high accuracy and recall rates in classifying distracted driving actions. Comparative analysis indicated that the custom-built dataset and the selected network architecture provided superior performance in terms of both accuracy and computation time compared to previous approaches. The lightweight nature of MobileNet allowed for efficient processing, meeting the real-time requirements necessary for in-vehicle monitoring. The significance of this work lies in its contribution to more robust and efficient driver monitoring systems. By combining facial feature analysis for fatigue with deep learning for distraction, the study offers a comprehensive solution that addresses two major safety risks simultaneously. The use of a lightweight CNN architecture highlights the feasibility of deploying such systems on embedded platforms within vehicles. The findings suggest that vision-based methods, when optimized with appropriate network architectures and curated datasets, can provide reliable, non-intrusive monitoring, thereby enhancing road safety and supporting the development of intelligent vehicle systems.

Key finding

Combining FAN-derived facial landmarks for PERCLOS-style fatigue estimation with a MobileNet CNN for distraction classification yields better accuracy and inference speed than prior single-purpose detectors when run on a webcam-equipped PC.

Methodology

other

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed. Discovered via discover_arxiv on 2026-05-04 (3 acquisition events logged).

Summary generated by qwen3.6-27b-prismaquant on 2026-06-10; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• distraction detection algorithms
• drowsiness detection algorithms

Information type

What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).

• Methodological Resource: tool software
• Theoretical Contribution: computational model, conceptual framework