Occupational Health and Safety (OHS) is a crucial aspect of industrial warehouse operations that involve high workplace risk, particularly regarding compliance with the use of Personal Protective Equipment (PPE). This study aims to develop an early warning system for OHS violations through the integration of smart cameras based on Convolutional Neural Networks (CNN) and the You Only Look Once (YOLOv8) algorithm to automatically and real-time detect PPE non- compliance in the Company XYZ. The research employed a Research and Development (R&D) method, consisting of needs analysis, system design, program coding, testing, as well as system implementation and maintenance. The detection objects include compliant and non-compliant PPE classes (hardhat, mask, and safety vest) and neutral objects, namely persons, safety cones, vehicles, and machinery. Experimental results show a significant improvement in model performance, with precision increasing from approximately 0.6 to 0.9, recall from 0.3 to 0.7, mAP@0.5 from 0.3 to 0.8, and mAP@0.5–0.95 from 0.1 to 0.5 at the 100th epoch. These results indicate that the proposed system can reliably detect PPE violations under dynamic working conditions and support the formulation of measurable internal OHS regulations and sanction mechanisms.