Driver Safety Development Real Time Driver Drowsiness Detection System Based on Convolutional Neural Network
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 paper addresses the critical issue of driver safety by proposing a real-time driver drowsiness detection system based on Convolutional Neural Networks (CNN). Motivated by the high prevalence of traffic accidents caused by driver fatigue, the authors focus on facial feature-based detection, specifically eye closure, as a non-invasive and cost-effective alternative to vehicle-based or signal-based methods. The primary challenge identified is the lack of comprehensive datasets for eye closure detection, particularly those accounting for varied lighting conditions and oblique viewing angles. To address this, the study introduces a novel framework that combines a new dataset with three distinct neural network architectures to achieve high accuracy and low computational complexity suitable for real-time applications. The methodology involves a preprocessing pipeline that detects the driver’s head using the Viola-Jones algorithm and identifies eye landmarks via a regression tree approach. The system crops the eye region of interest (ROI), converts it to grayscale, applies histogram equalization to mitigate lighting variations, and downsamples the image to 24x24 pixels to reduce computational load. The authors developed a new dataset containing 4,157 images from four individuals, featuring straight and oblique head views, various glasses types, and different ethnicities, which was combined with the existing ZJU dataset. Three networks were evaluated: a Fully Designed Neural Network (FD-NN) and two transfer learning models using pre-trained VGG16 and VGG19 architectures (TL-VGG). The system triggers an alarm if the network classifies the eye as closed for 12 successive frames (approximately 2 seconds at 6 frames per second), distinguishing drowsiness from normal blinking. Experimental results demonstrate that the FD-NN achieved the highest performance on the ZJU dataset, with an accuracy of 98.15% and an Area Under Curve (AUC) of 99.8%, outperforming the TL-VGG16 (95.45% accuracy) and TL-VGG19 (94.96% accuracy). On the extended dataset, TL-VGG16 performed best with 97.54% test accuracy, suggesting that deeper networks benefit from larger, more diverse datasets. Crucially, the FD-NN exhibited significantly lower computational complexity, requiring only 1.4 ms for eye state classification. This speed is approximately four times faster than comparable deep learning methods and twenty times faster than feature-extraction-based approaches, making it highly suitable for real-time implementation. The significance of this work lies in its demonstration that a lightweight, fully designed CNN can achieve superior accuracy and speed compared to deeper transfer learning models when trained on smaller datasets. The proposed system effectively balances the trade-off between model complexity and dataset size, offering a robust solution for driver monitoring systems. The inclusion of oblique views and diverse ethnicities in the dataset enhances the system's applicability in real-world driving conditions. The authors conclude that while binary classification of sleep states is effective, future work should explore multi-level drowsiness detection and yawning analysis to further improve system reliability.
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
simulator
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).
| Stage | Outcome | Tool | Model | Prompt | Attempts | Completed |
|---|---|---|---|---|---|---|
| discover | success | arxiv | — | — | 3 | 2026-05-04 |
| archive | success | — | — | — | 1 | 2026-05-04 |
| extract | success | cached | — | — | 2 | 2026-06-10 |
| clean | success | — | — | — | 1 | 2026-06-01 |
| chunk | success | — | — | — | 1 | 2026-06-01 |
| embed | success | — | — | — | 1 | 2026-06-02 |
| enrich | success | — | — | — | 1 | 2026-05-04 |
| promote | success | — | — | — | 1 | 2026-05-04 |
| summarize | success | llm | qwen3.6-27b-prismaquant | summ-v5 | 2 | 2026-06-10 |
| tag | success | vector_similarity | — | — | 16 | 2026-06-11 |
| verify | partial | — | — | — | 2 | 2026-06-10 |
Summary generated by qwen3.6-27b-prismaquant on 2026-06-10; verification: verified_with_issues.
Topics
Ranked by relevance to this paper. Hover a topic for its definition.
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