Real Time Detection of Safety Glasses using Machine Learning and Internet of Things Technology

This research presents the development of a real-time safety glasses detection prototype system for automated workplace access control. This system integrates the Vision Transformer (ViT) architecture, utilizing machine learning to classify glasses from webcam images, with an ESP32 board and Internet of Things (IoT) to control an electromagnetic lock. Initial test results after model development on an internal 60-image dataset showed 100% accuracy. When tested under real-world conditions using a new 120-image dataset, the model maintained high performance, achieving 98% overall accuracy. Notably, for the safety glasses class specifically, both Precision and Recall reached 100%. This demonstrates the capability to accurately detect safety glasses wearers, enabling effective control of workplace entry and exit.