AI-Based Real-Time Worker Safety Monitoring and Alert System

Construction and industrial work often involve hazards that can be reduced when personal protective equipment is used correctly and safety concerns are reported quickly. Manual observation remains important, but a supervisor may not be able to watch every worker continuously. This paper presents a prototype for real-time worker safety monitoring using computer vision. The system uses a custom-trained You Only Look Once (YOLO) object-detection model with a webcam feed to identify workers and visible personal protective equipment categories, including helmets, goggles, gloves, boots and vests, as well as selected non-compliance classes. OpenCV is used to process camera frames and display detection boxes and a safety-status panel. When configured violation classes are detected, the prototype displays an on-screen warning and produces a voice alert. An SMS notification workflow is also demonstrated through a phone notification screenshot supplied from the project test. The current evidence confirms that the application can display selected safety alerts and that an SMS-style notification was received during the reported demonstration. However, this paper does not claim a measured detection accuracy, a validated reduction in accidents or a measured SMS delivery time because a controlled quantitative evaluation has not yet been documented. The prototype is intended to support, rather than replace, trained safety personnel. Its practical value and limitations are discussed, together with the testing required before deployment in a real worksite.