Manuscript received November 17, 2025; revised December 17, 2025; accepted January 9, 2026; published April 28, 2026.

Abstract—Driver drowsiness is a major contributor to road accidents, creating an urgent need for real-time and nonintrusive monitoring systems that operate reliably under practical driving conditions. However, existing vision-based deep learning approaches often suffer performance degradation under illumination variation, facial occlusion, and transient visual noise, limiting their ability to capture subtle temporal manifestations of fatigue during gradual microsleep progression. This study proposes a robust temporal modeling framework for real-time driver drowsiness detection on Internet-of-Things (IoT) edge platforms based on a Domain-Aware Spatial–Temporal Selective-Gated Long Short-Term Memory (SG-LSTM) architecture. In the proposed framework, deep spatial features extracted by ResNet50 are fused with physiologically validated indicators, namely the Eye Aspect Ratio (EAR), Mouth Aspect Ratio (MAR), and head pitch angle, forming a unified 2051-dimensional temporal sequence representation. Two complementary gating mechanisms are introduced to suppress redundant spatial features and emphasize informative time steps during temporal reasoning. Experimental evaluation on the National Tsing Hua University Drowsy Driver Detection (NTHU-DDD) dataset, supported by real driving recordings, demonstrates that the proposed method achieves an accuracy of 90.09% and an F1-score of 0.90, outperforming Convolutional Neural Network–Long Short-Term Memory (CNN-LSTM) and vanilla Long Short-Term Memory (LSTM) baselines while maintaining robustness under realworld visual degradation. Deployment on an Orange Pi 3 edge device further confirms the feasibility of real-time inference on resource-constrained hardware. These results indicate that integrating domain-aware physiological cues with selective temporal gating provides an effective and practical solution for real-time driver drowsiness detection in intelligent transportation systems and IoT-edge environments.

Keywords—domain-aware physiological fusion, selectivegated LSTM, spatio-temporal micro-expression analysis, real-time embedded driver monitoring, vision-based drowsiness detection

Cite: Novriadi Antonius Siagian, Poltak Sihombing, Amalia, and Ade Candra, "Domain-Aware Spatial-Temporal Selective Gated LSTM for Real-Time Driver Drowsiness Detection," Journal of Image and Graphics, Vol. 14, No. 2, pp. 338-352, 2026.

Copyright © 2026 by the authors. This is an open access article distributed under the Creative Commons Attribution License (CC-BY-4.0), which permits use, distribution and reproduction in any medium, provided that the article is properly cited, the use is non-commercial and no modifications or adaptations are made.