EUVIMEDEUVIMED entdecken
Crossref

Sentry Eye: Real-Time Eye Strain and Fatigue Monitoring System using Deep Learning and Computer Vision

Salman Khan A, Dilshad Begum, Moinuddin S K

International Journal of Modern Science and Research Technology · 2026 · 4 (7) · 25-34

DOI: 10.67317/ijmsrt26jul025 ↗

Kurz gefasst

Fachartikel in der Zeitschrift International Journal of Modern Science and Research Technology, 2026. Verfasst von Salman Khan A und 2 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Prolonged interaction with digital displays is a leading cause of Digital Eye Strain (DES), yet continuous, affordable, and non-intrusive ocular fatigue assessment remains largely unavailable outside clinical settings.This paper presents SentryEye, a real-time eye strain and fatigue monitoring framework that integrates MediaPipe facial landmark extraction, adaptive Eye Aspect Ratio (EAR) analysis, PERCLOS quantification, deep learning-based strain classification, gaze estimation, and HSVbased eye redness scoring within a unified Flask-Socket.IO web architecture.Webcam frames are processed at 25-30 FPS entirely on standard CPU hardware; a 60-second adaptive calibration phase personalizes blink-detection thresholds, and a seven-condition alert engine with per-condition cooldown intervals delivers synchronized audio-visual notifications.Experimental validation on a three-class eye image dataset (Normal, Mild Strain, High Strain) partitioned in an 80:20 training-tovalidation ratio shows that the optimized MobileNetV2 transfer-learning classifier attains 92.3% validation accuracy, a weighted F1-score of 91.9%, a ROC-AUC of 0.971, and approximately 12 ms per-frame CPU inference, outperforming a custom EyeNet CNN baseline.The end-to-end pipeline sustains sub-50 ms Socket.IO metric emission latency and an average alert detection accuracy of 95.9% across all seven monitored conditions, confirming the deployability of the framework for institutional, workplace, and remote digital-wellness monitoring. Novelty in the ResearchSentryEye advances prior art on four fronts.First, it introduces an integrated, modular five-layer lightweight pipeline that executes facial mesh tracking, EAR/PERCLOS computation, CNN classification, temporal prediction, and telemetry emission concurrently at 25-30 FPS on a standalone CPU, without GPU acceleration or cloud offloading.Second, it embeds an adaptive personalized 60second baseline EAR calibration mechanism that derives user-specific closure thresholds from the top-90% sorted sample distribution, reducing false blink detections by approximately 23% relative to fixed global thresholds.Third, it deploys a multi-condition alert engine that manages seven synchronous physiological and behavioural factors-multi-face presence, prolonged eye closure, sustained fatigue, low blink rate, high PERCLOS, 20-20-20 screen-time compliance, and HSV-based eye redness-coupled with programmatic native sound synthesis through the Web Audio API, eliminating external audio dependencies.Fourth, it provides real-time interactive local reporting via a Flask-Socket.IO-Chart.jsdashboard with sub-50 ms emission latency, delivering complete analytics on-device

Zitieren

APA

A, S. K., Begum, D., & K, M. S. (2026). Sentry Eye: Real-Time Eye Strain and Fatigue Monitoring System using Deep Learning and Computer Vision. International Journal of Modern Science and Research Technology, 4(7), 25-34. https://doi.org/10.67317/ijmsrt26jul025

Vancouver

A SK, Begum D, K MS. Sentry Eye: Real-Time Eye Strain and Fatigue Monitoring System using Deep Learning and Computer Vision. International Journal of Modern Science and Research Technology. 2026;4(7):25-34. doi:10.67317/ijmsrt26jul025

BibTeX
@article{A2026,
  author = {A, Salman Khan and Begum, Dilshad and K, Moinuddin S},
  title = {Sentry Eye: Real-Time Eye Strain and Fatigue Monitoring System using Deep Learning and Computer Vision},
  journal = {International Journal of Modern Science and Research Technology},
  year = {2026},
  volume = {4},
  number = {7},
  pages = {25--34},
  doi = {10.67317/ijmsrt26jul025}
}
Kostenlos starten →