Fast Robust Large-scale Mapping from Video and Internet Photo ...

Keywords:
1. Introduction
The fully automatic modeling of large-scale environments has been a research
goal in photogrammetry and computer vision since a long time. Detailed
3D models automatically acquired from the real world have many uses
including civil and military planning, mapping, virtual tourism, games, and
movies. In this paper we present a system approaching the fully automatic
modeling of large-scale environments, either from video or from photo collections.
Our system has been designed for efficiency and scalability. GPU implementations
for SIFT feature matching, KLT feature tracking, gist feature
extraction and multi-view stereo allow our system to rapidly process large
amounts of video and photographs. For large photo collections, iconic image
clustering allows the dataset to be broken into small closely related parts.
The parts are then processed and merged, allowing tens of thousands of photos
to be registered. Similarly, for videos, loop detection finds intersections in
the camera path, which reduces drift for long sequences and allows multiple
videos to be registered.
Recently mapping systems like Microsoft Bing Maps and Google Earth
have started to use 3D models of cities in their visualizations. Currently these
systems still require a human in the loop for delivering models of reasonable
quality. Nevertheless they already achieve impressive results, modeling large
areas with regular updates. However, these models are very low complexity,
and do not provide enough detail for ground-level viewing. Furthermore the
2