Point Cloud Segmentation in Urban Areas

3D segmentation methods can be applied to the segmentation studies of large scale point cloud data in urban areas. In this study, point cloud segmentation was carried out for three big Chinese cities, nanemly Chengdu, Jiaoda, Shenzhen. Due to not having enough labelled data from these urban areas for model training, transfer learning techniques were adapted using SensatUrban and Toronto-3D datasets. Experimental results show that proposed method enables achieving F1 scores above 80% for each region. Transfer learning strategy offers a reliable solution to overcome the problem of labelled dataset scarcity for the task of 3D point cloud segmentation in urban areas. The findings also contribute to the advancement of point cloud segmentation methods, particularly in the context of rapidly developing Chinese urban areas.