August 19, 2025Open Access

Robust Object Detection in Industrial Safety Environments Using YOLOv8

HSHani SirajNational University of Computer and Emerging Sciences

Key Points

A final mAP@0.5 score of 85.1% demonstrates effective object detection of critical categories like fire extinguishers.
The system was fine-tuned on 1,139 annotated images, addressing challenges like occlusion and lighting sensitivity.
Optimizations included data augmentation and hyperparameter tuning, enhancing the model's resilience in real-world applications.
This approach highlights the potential of deploying lightweight object detection models for increased safety in industrial settings.

Abstract

Abstract In industrial and safety-critical environments, real-time object detection plays a crucial role in hazard prevention and operational efficiency. This shared task presents a high-performance object detection pipeline using YOLOv8 to identify three critical object categories: Fire Extinguishers, Toolboxes, and Oxygen Tanks. The system was trained and fine-tuned on a dataset of 1,139 annotated images with a balanced class distribution. The project focused on optimizing model performance through data augmentation, hyperparameter tuning, and occlusion handling. A final mAP@0.5 score of 85.1\% was achieved. The study addresses key challenges such as overfitting, occlusion-based misclassification, and lighting sensitivity. This work demonstrates an effective approach to deploying lightweight yet accurate object detection models in real-world industrial settings.

Demander à l'IA

Bookmark

View Full Paper