Real-Time Driver-Drowsiness Detection System Using Facial Features

Deng, Wanghua; Wu, Ruoxue · 2019 · DOAJ

DOI: 10.1109/ACCESS.2019.2936663

archive: archived pipeline: cataloged verified

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

Summary

This paper addresses the critical safety issue of fatigue-induced traffic accidents, which account for approximately 20–30% of incidents in the United States. To mitigate this risk, the authors propose DriCare, a non-contact, real-time driver-drowsiness detection system that utilizes video images from a vehicle-mounted camera. Unlike subjective methods requiring driver input or contact-based systems requiring wearable sensors, DriCare analyzes facial features—specifically eye closure and yawning—to assess fatigue levels without imposing additional burdens on the driver. The system aims to overcome limitations in existing algorithms, such as poor tracking accuracy in complex lighting conditions and the inability to automatically initialize tracking or recover lost targets. The methodology centers on a novel face-tracking algorithm called Multiple Convolutional Neural Networks-KCF (MC-KCF). This algorithm integrates the Kernelized Correlation Filter (KCF) with features extracted from a SqueezeNet convolutional neural network (CNN) and Felzenszwalb Histogram of Oriented Gradients (FHOG). By fusing these features with optimized weights, MC-KCF improves tracking stability in low-light or complex environments. To address initialization and target loss, the system employs Multi-task CNN (MTCNN) to periodically calibrate the tracker, specifically at the start of a session or every 10 seconds. Once the face is tracked, the system uses Dlib to locate 68 facial key points. It then defines specific regions for the eyes and mouth using geometric calculations based on these landmarks. The state of the eyes is evaluated using a CNN that measures the angle of eye opening, while yawning is detected by assessing the duration of mouth opening. The system evaluates driver drowsiness based on three criteria: blinking frequency, duration of eye closure, and yawning frequency. If these metrics exceed predefined thresholds, the cloud server processes the data and sends a warning alert to the driver’s mobile device. The experimental results demonstrate that DriCare achieves an accuracy of approximately 92%. The integration of CNN features into the correlation filter framework significantly enhances tracking performance compared to the original KCF algorithm, particularly in maintaining accuracy during head movements and varying illumination. The significance of this work lies in providing a cost-effective, non-intrusive solution for real-time fatigue detection that can be widely deployed in vehicles. By combining deep learning with efficient correlation filters, the system balances computational speed with high accuracy, making it suitable for real-time applications. The proposed MC-KCF algorithm and the specific facial region evaluation methods offer a robust alternative to existing contact-based or less accurate non-contact systems, contributing to the advancement of intelligent transportation safety technologies.

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.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified_with_issues.

Topics

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

• drowsiness detection algorithms
• drowsiness
• gaze based attention detection

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).

• Empirical Findings: physiological data
• Methodological Resource: tool software