<![CDATA[This study aims to develop SentryNap, a real-time drowsiness warning and monitoring system for industrial control-room operators based on YOLOv5s using CCTV/webcam image input. Reduced alertness caused by long shift work and operator drowsiness can increase the risk of operational errors, while many previous approaches still rely on intrusive physiological sensors or visual methods that are sensitive to changes in lighting and head pose. The proposed system uses a YOLOv5s model fine-tuned on a Roboflow dataset with two classes, namely normal and sleeping, and integrates the inference results with a Node.js backend for JSON logging. Model training was conducted for 50 epochs at a resolution of 640 x 640 pixels using the SGD optimizer, while evaluation was carried out through static validation and real-time testing scenarios. The model achieved a precision of 0.986, recall of 1.000, mAP@0.5 of 0.995, and mAP@0.5:0.95 of 0.621. Real-time testing showed that detection results could be recorded by the backend in less than one second. These findings indicate that SentryNap has potential as a non-invasive operator safety monitoring prototype, although larger datasets and broader field validation are still required.]]>
Copyrights © 2026
