Aerial Terrain Mapping
3D model building for urban and other environments is an increasingly important technology. This project is the first-stage component of a system that builds 3D models from aerial images. The goal is to compute camera position for aerial images followed by 3D features for structures and terrain within the images.
Background & Objective: 3D urban models support everything from online maps and orientation tools, to live augmentation of video of an urban area, to government planning for urban infrastructure, to disaster recovery. Existing model-creation tools often require manual support which is time-consuming. Our goals are to automate the 3D model construction as far as possible and to support real-time operation.
Technical Discussion: This system uses a traditional approach to the problem - computing point features in the aerial images, matching them, computing the camera positions, then reconstructing the 3D point positions. The processing proceeds in an iterative fashion, starting with no initial estimate of the 3D scene, then constraining the matching at each subsequent iteration such that points lie on a terrain-like surface and outlying points are eliminated. The software accepts various starting conditions such as unknown/known camera position, and allows the user to provide hints about the position of the terrain surface if desired.
Future Direction: The initial work was on computing 3D points only. This can be extended by computing 3D lines, which are as common as points in urban environments.
Contacts:
Paul Beardsley
Jeroen van Baar
Technology Areas:
Imaging
Computer Vision
Modification Date: July 3, 2007
