Pothole and Crack Detection on Asphalt Pavement in Bandung through Machine Learning Approach Rifdah Puspita Sari (a), Yackob Astor (b*), Iwan Awaludin (c), Salman Shalahuddin (d)
a) Department of Civil Engineering, Politeknik Negeri Bandung, Jl. Gegerkalong Hilir, Kabupaten Bandung Barat 40559, Indonesia
b) Department of Civil Engineering, Politeknik Negeri Bandung, Jl. Gegerkalong Hilir, Kabupaten Bandung Barat 40559, Indonesia
*yackobastor[at]polban.ac.id
c) Department of Computer and Informatics Engineering, Politeknik Negeri Bandung, Jl. Gegerkalong Hilir, Kabupaten Bandung Barat 40559, Indonesia
d) Independent Researcher, Bandung, Indonesia
Abstract
Road condition surveys are typically performed manually. Surveyors walk along the road directly to identify and measure the type of damage. This approach is time-consuming and labor-intensive. With advancements in Artificial Intelligence, it is possible to detect road damage such as potholes and cracks, hence improving the efficiency of data processing in road condition surveys. This research aims to semi-automatically detect asphalt road damage through image data processing using a machine learning approach. The YOLOv8n model was used for road damage detection, achieving a highest mAP50 of 0.36. Therefore, the model was used to detect potholes and cracks in Terusan Jakarta Road. As a result, the Surface Distress Index (SDI) score for 1 km road segment of Terusan Jakarta Road based on the detection was 17, indicating that the road is in good condition and requires routine maintenance.
