Building facade extraction from roadside images for 3D Building models

초록

This study investigating how to create building facades from roadside images taken by conventional camera and to map them onto the existing 3D building models. In our previous study, we had generated a technique to create 3d models of buildings by using shadow analysis from single-view images. This technology can work with satellite/aerial images without metadata, including Google earth images. It can create 3D wireframe model for a building with texture information on a building roof. However image resolution and building distribution restricted the accuracy of 3D building model. Moreover, mapping building facades is also a problem to solve. This paper tries to introduce the study that presently conducted to extend previous research. We propose techniques for generating and mapping building facades onto 3D building models. In the study, simple 3D models were creating from Google earth images using the technology developed in our previous study. The coordinates of building foot corners were found from the images, and their UTM coordinates were calculated. We took photos of building facades along the roadside with a conventional digital camera. When taking roadside photos, camera position and approximate orientation were recorded. Based on the relationship between camera orientation and 3D building models, the boundaries of 3D building models were drawn onto the roadside images. On the other side, straight line features were extracted from the photos. The boundaries of 3D building models were matched against these line features for finding correct image position of building facade models and to extract building facade information. Finally, the study tested the possibility of updating 3D building models using line features of the roadside photos. This study provides a simple and cost-effective method for creating building models with texture information on building roof and facades.