Real-time personal protective equipment compliance detection using you only look once

Nurhanisah, Aminuddin and Nor Azuana, Ramli and Pratondo, Agus (2024) Real-time personal protective equipment compliance detection using you only look once. In: 2024 5th International Conference on Artificial Intelligence and Data Sciences, AiDAS 2024 - Proceedings. 5th International Conference on Artificial Intelligence and Data Sciences, AiDAS 2024 , 3 - 4 September 2024 , Hybrid, Bangkok. pp. 292-297.. ISBN 979-833152855-3 (Published)

Pdf Real-Time_Personal_Protective_Equipment_Compliance_Detection_Using_You_Only_Look_Once.pdf Restricted to Repository staff only Download (402kB) | Request a copy

Abstract

Developing automated systems to detect and identify protective equipment in working sites is not just a technical accomplishment but also a significant step towards improving the workplace, especially in high-risk areas like construction and manufacturing sites. The objective of detecting safety equipment is to identify and classify the equipment worn by workers to ensure compliance with safety standards. The proposed research work employs the system of object detection from CCTV videos around the working sites of company XYZ through a framework that includes image acquisition, data collection, and video data pre-processing, with the implementation of various deep learning algorithms using state-of-the-art models such as You Only Look Once (YOLO) including YOLOv5, YOLOv7, and YOLOv8. The primary dataset, shared through a cloud server for confidentiality, is used for training. The model's performance is evaluated using metrics such as confusion matrix, accuracy, precision, recall, mean average precision (mAP), and losses. Results indicate that YOLOv8 exceeds the performance of the other YOLO models, achieving precision, recall, and mAP of over 90% for all identified classes: HELMET, NOHELMET, SAFETYJACKET, NOSAFETYJACKET. Further analysis comparing the selected models based on the confusion matrix shows that YOLOv8 demonstrates more accurate predictions, while YOLOv5 and YOLOv7 primarily detect background, leading to minimal object detection within the bounding boxes.

Item Type:	Conference or Workshop Item (Paper)
Additional Information:	Indexed by Scopus
Uncontrolled Keywords:	image processing, object detection, safety equipment, smart surveillance, You Only Look Once
Subjects:	Q Science > QA Mathematics > QA76 Computer software
Faculty/Division:	Institute of Postgraduate Studies Center for Mathematical Science
Depositing User:	Dr. Nor Azuana Ramli
Date Deposited:	06 Jan 2025 04:45
Last Modified:	06 Jan 2025 04:45
URI:	http://umpir.ump.edu.my/id/eprint/43457
Download Statistic:	View Download Statistics

Actions (login required)

View Item