CCTV Integration of YOLOv8 for Human Detection and Safety Enhancement in Forklift Work Environments

Abstract

Workplace accidents caused by collisions between workers and forklifts represent one of the most significant safety risks in manufacturing environments. Limited workspace often forces forklift operational paths to intersect with pedestrian walkways, creating hazardous zones that require proactive mitigation strategies. This study investigates the implementation of the You Only Look Once version 8 (YOLOv8) method on existing CCTV systems for real-time human detection as an effort to reduce forklift-related incidents. The methodology includes the development of an optimized YOLOv8 architecture tailored for industrial settings, model training using Roboflow with a dataset consisting of photos collected directly from the factory area, and integration with the facility’s speed-door system that governs forklift entry and exit. Detection performance testing demonstrated that the model achieved 100% accuracy across 55 test images, with precision, recall, and F1-score each reaching a perfect value of 1. These findings indicate that YOLOv8 is highly reliable for real-time person detection in industrial environments and can be effectively deployed using existing CCTV infrastructure. It is expected that workplace accidents can be minimized by preventing the door-open mechanism activated when the CCTV system detects the presence of a person. Integrating the detection system with automated door control offers a promising safety enhancement, helping to restrict unsafe access to forklift pathways and reduce the risk of collisions in manufacturing settings.