Andy Desman, Kahlil Muchtar, Maulisa Oktiana, Maya Fitria, Yuwaldi Away, Fardian Fardian, Hammam Riza
The WHO predicts that by 2030 road accidents will be the 5th leading cause of death. Globally, road accidents account for 1.25 million casualties each year, and road defects cause 34% of these casualties. The road survey process in many countries have several challenges, one of which is detection using cameras that do not have a recognition system. In this study, a model with YOLOS architecture based on Vision Transformer trained on the RDD2022 dataset successfully recognizes road damage well, as indicated by the number of objects detected, bounding box on accurate objects, and the ability to recognize objects with inconsistent shadow and light inference. This research uses assessment parameters such as Average Precision (AP) and Average Recall (AR) to determine the overall performance of the model. The model achieves the highest AP value at Intersection of Union (IoU) 0.5, 0.75, and 0.5-0.95, worth 62.1%, 37.1%, and 36.2% respectively, and the highest AR value in Large, Medium, and Small Areas, worth 42.1%, 60.3%, and 75.4% respectively. The supplementary material can be found through this link: https://www.youtube.com/watch?v=LzkI2e_IORE. © 2025 IEEE.
Dept. of Electrical and Computer Engineering, Universitas Syiah Kuala, Banda Aceh, Indonesia; Dept. of Electrical and Computer Engineering, Universitas Syiah Kuala, The Master Program in Artificial Intelligence, The Department of Informatics, Universitas Syiah Kuala (USK), Banda Aceh, Indonesia; BRIN (National Research and Innovation Agency), Indonesia The Master Program in Artificial Intelligence, The Department of Informatics, Universitas Syiah Kuala, Banda Aceh, Indonesia
Research at a Glance
Register to unlock
Topics & SDG Alignment
Register to unlock
Collaboration
Register to unlock
Author Profile (Selected)
Register to unlock
References Overview
Register to unlock
Journal & Source
Register to unlock
Metadata & Integrity
Register to unlock