An effective approach for real-time drowsy driving prediction using quantized fisher-Gabor features and latent-dynamic conditional random fields

Bakheet, Samy; Al-Hamadi, Ayoub; Alanazi, Abed · 2025 · Crossref

DOI: 10.3389/fcomp.2025.1437084

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 drowsiness, a leading cause of traffic accidents responsible for approximately 20% of road collisions. The authors propose an automated, vision-based system for real-time prediction of driver fatigue, aiming to provide timely warnings to prevent accidents. The research is motivated by the limitations of existing physiological sensors, which are invasive and impractical for production vehicles, and the need for robust, efficient computer vision solutions that can operate under varying lighting and viewing conditions. The proposed methodology employs a multi-stage pipeline. First, input images from a dashboard camera are preprocessed using a 2D Gaussian blur filter to reduce noise and an adaptive contrast-limited histogram equalization technique to compensate for illumination variations. Facial landmarks are then localized using an AdaBoost classifier based on Haar-like features, followed by an enhanced Active Shape Model (ASM) to precisely identify regions of interest, such as the eye and mouth areas. For feature extraction, the system utilizes a novel Fisher-Gabor Descriptor (FGD). This involves convolving facial regions with a bank of 40 Gabor filters (eight orientations and five scales) to capture local texture features. To ensure robustness against scale, rotation, and illumination changes, and to reduce computational redundancy, the authors apply Fisher’s quantum information theory to calculate non-extensive entropies and Fisher information measures, selecting the most discriminative features. Finally, these normalized FGDs are fed into a Latent Dynamic Conditional Random Field (LDCRF) classification model, which predicts the driver’s state as either drowsy or awake. Experiments were conducted on the benchmark NTHU-DDD video dataset to evaluate the system’s performance. The results demonstrate that the proposed approach achieves a competitive detection accuracy of 97.6%, outperforming several state-of-the-art alternatives. Crucially, the system maintains stringent real-time processing guarantees, making it suitable for practical deployment in intelligent transportation systems. The LDCRF model proved effective in capturing the dynamic dependencies between visual observations and class labels, contributing to the high accuracy. The significance of this work lies in its contribution to non-invasive, real-time driver monitoring systems. By combining robust feature extraction via Fisher-Gabor descriptors with the predictive power of LDCRFs, the method offers a reliable solution for detecting fatigue under challenging environmental conditions. This approach supports the development of safer intelligent transportation systems by providing an effective, automated warning mechanism that can help reduce the incidence of drowsy-driving-related accidents.

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.

Topics

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

• drowsiness detection algorithms
• gaze based attention detection
• distraction 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).

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