Development of a Low-Cost Computer Vision-Based Safety Compliance System for Accident Prevention in Engineering Workshops in Polytechnics in South-West Nigeria

Abstract

Engineering workshops are hazardous, and personal protective equipment (PPE) compliance is hard to enforce because supervisors cannot observe every worker continuously. This study developed and evaluated a low-cost computer vision-based safety compliance system for accident prevention in engineering workshops in polytechnics in South-West Nigeria. A design-and-development experimental approach combined a workshop safety assessment with system development and field testing. A YOLOv8 object-detection model was trained on 4,200 workshop images augmented with open PPE datasets to detect five PPE classes (helmet, goggles, gloves, protective clothing and safety boots), and was deployed on a Raspberry Pi 4 edge device with low-cost cameras and real-time alerts. The model achieved precision of 0.86, recall of 0.82 and mAP@50 of 0.87. Field evaluation showed a detection accuracy of 89.6%, inference speed of 22.4 frames per second, end-to-end latency of 1.8 seconds and uptime of 97.4%. Expert validation rated the system highly for functionality (3.60), reliability (3.40) and affordability (3.80). Low-cost computer-vision monitoring is technically feasible and practically valuable for accident prevention in Nigerian polytechnic workshops, and the study provides a deployable design for wider adoption.