Driver Safety Development Real Time Driver Drowsiness Detection System Based on Convolutional Neural Network
URL: http://arxiv.org/abs/2001.05137v3
archive: archived pipeline: cataloged
Abstract
This paper focuses on the challenge of driver safety on the road and presents a novel system for driver drowsiness detection. In this system, to detect the falling sleep state of the driver as the sign of drowsiness, Convolutional Neural Networks (CNN) are used with regarding the two goals of real-time application, including high accuracy and fastness. Three networks introduced as a potential network for eye status classifcation in which one of them is a Fully Designed Neural Network (FD-NN) and others use Transfer Learning in VGG16 and VGG19 with extra designed layers (TL-VGG). Lack of an available and accurate eye dataset strongly feels in the area of eye closure detection. Therefore, a new comprehensive dataset proposed. The experimental results show the high accuracy and low computational complexity of the eye closure estimation and the ability of the proposed framework on drowsiness detection.
Summary
Methodology paper proposing a real-time driver drowsiness detection system using Convolutional Neural Networks for eye-closure classification. Three networks compared: a Fully Designed Neural Network (FD-NN) and two Transfer Learning networks based on VGG16 and VGG19 (TL-VGG16, TL-VGG19) with added fully connected layers. A pipeline extracts the eye region of interest via facial landmarks, converts to grayscale, equalizes contrast, and resizes to 24x24 pixels before classification. An alarm is triggered when eyes remain closed for 12 successive frames (~2 s at 6 fps). The authors also introduce an extended eye-state dataset adding an oblique-view category to supplement the ZJU dataset. No human-subject driving experiment; evaluation is on static eye-image datasets.
Key finding
On the ZJU eye dataset, FD-NN achieves 98.15% accuracy and 99.8% AUC for open/closed eye classification with 1.4 ms inference time per image, outperforming the heavier TL-VGG16 and TL-VGG19 transfer-learning variants on the speed-accuracy trade-off and supporting real-time drowsiness detection.
Methodology
Computer-vision benchmark study. Three CNN architectures (FD-NN, TL-VGG16, TL-VGG19) trained and evaluated on the ZJU eye dataset and an extended dataset compiled by the authors that adds an oblique-view eye class. Hardware: Intel Core i7-6700K, 16 GB RAM, NVIDIA GTX 1070Ti; Python/Anaconda. Metrics: classification accuracy, AUC, and per-image inference time. Drowsiness decision rule: 12 consecutive closed-eye frames at 6 fps (~2 s) triggers an alarm. No on-road or simulator driving data; no human-subject driving behavior measures.
Quality score: 5 / 5