Procedural Generation of 3D Scenes for Urban Landscape Based on Remote Sensing Images

3D Reconstruction is a significant research topic in the fields of Computer Graphics, Remote Sensing, Virtual Reality and so on. Particularly, 3D reconstruction of cities is one of the core technologies in urban infrastructure, ecological protection and transportation management. However, City-level scenes are characterized by large-scale and high complexity. The existing large-scale 3D scene reconstruction technology exists problems such as long time and low accuracy. Therefore, it is particularly important to improve the efficiency and realism of large-scale 3D reconstruction of urban scenes. In this paper, a programmed generation of urban largescale 3D scenes based on remote sensing images is proposed. Firstly, semantic-level instance segmentation information is extracted from the urban landscape objects in remote sensing images, then 2D contour information is extracted according to the segmentation results for 3D reconstruction of contour lines, and finally, a highly realistic 3D scene of large-scale urban landscape is rapidly generated according to landscape design rules and Procedural Content Generation Framework (PCG) technology. The method in this paper combines the accurate spatial information in remote sensing images with PCG rapid generation, which can improve the generation speed of 3D scenes and increase the realism of 3D scenes. The experimental results show the effectiveness of this paper’s method.