You Only Look Once Version 11 (YOLOv11) Based Object Detection for 3D City Modeling: A Study in the Jatirejo Area

Abstract

Digital three-dimensional (3D) modeling has become an essential requirement in modern spatial mapping and visualization, as it can provide a more realistic and detailed representation of objects or areas. This study uses image segmentation techniques using YOLO v11, which automatically recognizes and separates objects, thereby increasing the accuracy of image extraction and accelerating the modeling process. With the help of software and the web, namely Roboflow, Google Colab, and QGIS. The results of this study show that the integration of image extraction algorithms, deep learning, and image segmentation based on YOLO v11 produces a more precise, efficient, and realistic 3D model. The Confusion Matrix shows that the segmentation results are perfectly detected at a rate of more than 85.64%, with the remaining segmentation not being perfectly detected, accounting for 9.23% and 5.13% of the undetected part. The calculation of the precision value of 95.94% indicates that the model rarely makes mistakes in predicting objects. The resulting Recall value is 87.58% and the F1 score is 91.57%. Thus, the use of AI-based technology and computer vision offers an innovative solution in accelerating the development of effective, accurate, and cost-effective digital 3D city models that can be used as local government data.