Akademik Veri Yönetim Sistemi
International Conference on Engineering, Natural and Social Sciences ICENSOS, Konya, Türkiye, 22 - 23 Ekim 2024, ss.50-56
Potholes are a significant concern for roadway safety and infrastructure integrity. They can lead to severe vehicle damage, accidents, and increased maintenance costs for municipalities. Detecting and addressing potholes promptly is crucial to ensuring safe driving conditions, reducing repair expenses,
and enhancing overall traffic flow. This study presents a pothole detection system using the YOLOv8m model, trained to achieve high accuracy in both object detection and classification. The dataset used in this study is sourced from the Kaggle platform, a well-known repository for diverse datasets in various
domains. The model demonstrated effective performance, achieving a precision of 0.88, recall of 0.62, mean Average Precision (mAP50) of 0.75, and mAP50-95 of 0.47. Additionally, the developed system is tested on the Jetson TX2 embedded platform, known for its powerful processing capabilities and portability, achieving operational speeds about 5.88s. These results underscore the model's effectiveness, demonstrating its suitability for integration into embedded systems and autonomous operations, which is essential for timely pothole detection and response, ultimately enhancing roadway safety and maintenance efficiency.