<![CDATA[The detection of traffic prohibition signs in tropical urban environments is under-documented, as existing benchmark datasets such as GTSRB and TT100K do not represent the specific conditions of Southeast Asia. This study evaluates YOLOv5s for detecting and classifying six classes of traffic prohibition signs on four urban roads in Bandar Lampung, Indonesia, using a dataset of 9,898 labeled images extracted from real-world video recordings under various environmental conditions. YOLOv5s was directly compared with YOLOv4, YOLOv5m, and Faster R-CNN under identical evaluation conditions. YOLOv5s outperformed all comparison models with an average accuracy of 93.34% and an average F1-Score of 95.97%, with performance ranging from 88.65% in Pagar Alam to 97.28% at Unila, reflecting the documented gradation of environmental complexity. Processing speeds of 7.3–8.8 FPS place the system in the near-real-time category, making it suitable for offline traffic monitoring applications. This study provides a method for detecting prohibition signs in tropical urban environments in Indonesia and offers a practical reference point for the development of intelligent transportation systems in developing cities facing similar environmental challenges.]]>
Copyrights © 2026
