An Image Query-Based Approach for Urban Modeling

Effective 3D modeling for urban landscape reconstruction is one of the most important fields in 3D Computer Graphics. There have been proposed many modeling techniques. Procedural modeling method [M¨uller et al. 2006] is powerful and effective, however, it reconstructs buildings one by one precisely. Each building has to be specified as a procedure of various kinds of construction parts of buildings. Therefore, it takes long time to reconstruct urban landscapes. We have been reconstructing cityscapes of downtown Tokyo from the late Edo era to present Tokyo for several years in collaboration with Edo-Tokyo Museum[Sakuma et al. 2008]. We have modeled more than 200 buildings and their facade textures are stored in a database.In our experience, bottle necks of urban modeling tasks are as follows:

•  Landmark buildings have to be reconstructed precisely.

• Buildings neighboring the landmarks are not required to be reconstructed precisely, however, their appearances are needed to be similar to actual buildings.

• If buildings are reconstructed one by one, many additional tasks would be caused. For examples, deformation of facade textures of buildings have to be removed and the textures have to be adjusted to 3D models. The height of 3D buildings have to be adjusted to adjacent buildings.

We noticed that most buildings are located along with streets, and there are many buildings whose facades are very similar. Based on this analysis, we propose a novel method to model several buildings from a photograph. Our method is based on image query techniques which search and find similar buildings from a facade texture database to the buildings in the photograph. In addition, our tool reconstructs several buildings simultaneously .

Images

Tags

Tag: Katsuhisa Kanazawa SIGGRAPH Asia Cyberscape 東京