Skip to content
SDGs Center UDINUS

Evidence record · UDINUS · SDGs

Comparison of YOLO and RF-DETR Performance in Detecting Personal Protective Equipment in Construction Environments Using Tokenization

Rektorat UDINUS 2025 Published 2026-10-30

Evidence record · UDINUS · SDGs

Unit
Rektorat UDINUS
Period
2025
Goals
Record
b7a47f0a-813c-4ddf-828a-e919ec3e847e
Published
2026-10-30

Universitas Dian Nuswantoro (UDINUS), through the participation of Buana, Chandra Lukita; Shidik, Guruh Fajar; Ubaidillah, Muhammad Naufal; Grafer, Yudha Rizqia; Kristiyan, Filipus Arif; and Kusuma, Edi Jaya, conducted research comparing object detection models for identifying Personal Protective Equipment (PPE) in construction environments. The activity addressed occupational safety in a setting where workers face various potential hazards. It focused on computer vision as an efficient approach to PPE detection and examined the use of tokenization for object detection based on text commands.

The research compared three current object detection models: YOLOv8, YOLOv12, and RF-DETR. The models were equipped with tokenization techniques that enabled the system to process image and text inputs simultaneously. The detection scope included PPE objects such as helmets, vests, and gloves. The activity examined the models' accuracy, flexibility, precision, specificity, F1 score, and mAP50 performance in identifying relevant objects.

The evaluation showed that RF-DETR excelled in precision and F1 score, while YOLOv12 provided the best results on mAP50. YOLOv8 lagged slightly behind in precision but remained effective in applications with limited resources. The findings illustrate that tokenization improved the models' ability to identify PPE items with greater precision and specificity. The integrated approach was described as a practical and efficient real-time solution for workplace safety monitoring. It supports more accurate monitoring and timely response in environments where safety is critical.

Through this activity, UDINUS contributed research on computer vision, tokenization, object detection, and workplace safety monitoring. The institutional framing remains grounded in the comparison of detection models and the identification of PPE in construction environments.

Reference https://www.scopus.com/pages/publications/105031574370?origin=resultslist