ABSTRACT
The reconstruction of buildings and landscapes in 3D from 2D images and videos has long been a topic of research in computer vision and photogrammetry. In recent years, applications have been very successful in delivering effective visualizations of large-scale models based on aerial and satellite imagery to a broad audience. Visualizations using such data are possible in the form of panoramic mosaics (Teller et al. 2003, Román et al. 2004) or simplifi ed geometric models (Cornelis et al. 2006), which require less data to be constructed but limit the user’s ability to freely navigate the environment. To enable unconstrained navigation in the virtual environment, geometrically accurate and detailed 3D models are required. In the last years, commercial vendors have begun acquiring ground-based videos of primarily cities in most parts of the world. The massive amounts of captured data pose tremendous challenges for the collection, processing, and visualization of these data. Until now no ubiquitous, automatically reconstructed, high-quality models have been delivered to the general public due to the lack of fast fully automatic reconstruction methods.
