Deep convolutional network based real time fatigue detection and drowsiness alertness system

Sharma, Vijay Prakash; Yadav, Jitendra Singh; Sharma, Vivek · 2022 · Crossref

DOI: 10.11591/ijece.v12i5.pp5493-5500

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 driver fatigue and drowsiness, which contribute significantly to road accidents and fatalities. Motivated by the limitations of existing Advanced Driver-Assistance Systems (ADAS)—such as the invasiveness of physiological sensors (EEG, heart rate) and the unreliability of vehicle-based metrics—the authors propose a non-intrusive, real-time detection system based on computer vision and deep learning. The goal is to create an efficient, lighting-invariant method for monitoring driver alertness through facial feature analysis, specifically focusing on eye state and yawning behavior. The proposed method employs a three-phase strategy utilizing a Convolutional Neural Network (CNN) and traditional image processing techniques. First, face detection is performed using the Viola-Jones algorithm. To ensure lighting invariance, the system converts images to the YCbCr color space, segmenting skin based on chromatic components to reject non-face backgrounds. Second, the system tracks eye states and detects yawning. Eye positioning is achieved via Viola-Jones, followed by edge detection using Sobel and Canny algorithms to determine eyelid height and openness. Yawning detection involves segmenting the mouth area and applying K-Means clustering with correlation coefficient template matching to distinguish between open and closed mouth states. Finally, feature vectors from these phases are concatenated and classified using a binary Support Vector Machine (SVM) classifier. The CNN architecture consists of seven layers, including two convolutional layers with 64 channels and ReLU activation, designed to process 640x480 resolution images at 30 frames per second. The system was evaluated across four distinct lighting scenarios: broad daylight, low light, artificial night light, and low-light close proximity. Results indicate that the system performs best in controlled or well-lit environments. In broad daylight (Scenario 1), the average accuracy was 91%, and in artificial night light (Scenario 3), it reached 93%. Performance dropped in low-light conditions (Scenario 2) to 81% and further to 68% in low-light close proximity (Scenario 4). Across all trials, the detection of yawning proved more accurate than the detection of eye movement as a sign of sleepiness. The system successfully triggers an alarm when the duration of the fatigue state exceeds a predefined threshold. The study concludes that the proposed framework offers a robust, real-time solution for drowsiness detection that is largely invariant to illumination changes, outperforming many previous approaches in accuracy. The integration of CNNs with traditional feature extraction methods allows for effective classification of fatigue states. However, the authors note limitations, including reduced accuracy when drivers wear glasses and the system's inability to distinguish between nodding off and head tilting. Future work aims to address these issues by incorporating head-lowering predictions and improving robustness against accessories like eyewear.

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-nvidia on 2026-08-10; verification: verified.

Topics

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

• drowsiness detection algorithms
• drowsiness