Skip to content
Conference Open access

An Enhanced Deep Learning Framework for Accurate Detection of Safety Helmet Wearing Compliance

Oct 2026 · International Conference on Innovative Mechanisms for Industry Applications · pp. 424-430 · 0 citations · 17 references

Abstract

Reducing injuries and deaths in construction, mining, and industrial sectors requires ensuring safety helmet compliance. Manual supervision is at the heart of traditional monitoring approaches; nevertheless, it is laborious, prone to mistakes, and ill-suited to modern, sometimes unpredictable, workplaces. This research introduces YOLO-ESCA, a YOLOv5-based system that takes use of Soft-NMS, CBAM, and an ADL to boost uncertainty classification and tiny object recognition. Three classes are identified by the model: helmet, head, and uncertainty. This allows for effective detection of correct and poor helmet use. Integrating a YOLOv11-based model with improved feature extraction, finer detection heads, and optimized learning algorithms significantly improves detection performance. Benchmarks on the SHWD dataset show that compared to the enhanced YOLOv5 model, the suggested YOLOv11 framework outperforms it in terms of mAP50-95, accuracy, and recall. Further, the system is compatible with video and image-based monitoring, so it can be used in real-world deployments to assist safety managers in automatically detecting non-compliant employees and improving safety procedures in a variety of industrial settings.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.