Multi-Image Road Extraction
Abstract
Our research is focused on an investigation of automated road tracking using multiple images, toward a goal of fully automated extraction of 3D road networks with topology and attribution. The use of multiple images for road tracking makes the process more robust, due to analysis of the scene from different view points. It also supports direct extraction of 3D information along the path of the road. Determination of road elevation has significant implications for reducing cost and time in applications requiring cartographic features with full 3D attribution. These include mission planning and rehearsal, visualization in urban areas, and the automated production of digital cartographic products. Under this ARO research contract a framework for multi-image road extraction was developed and implemented (RoadMAP3D) with an interactive user interface, tailored to simplify interactions. A detailed quantitative analysis of RoadMAP3D performance is derived and presented. This includes the development of two reference data sets with 3D road geometry, metrics for error calculation with respect to automatically generated road networks, and visualizations of the extracted roads using road height and digital elevation models.
Document Details
- Document Type
- Technical Report
- Publication Date
- Nov 22, 2005
- Accession Number
- ADA440760
Entities
People
- David M. Mckeown
- Steven D. Cochran
- W. A. Harvey
Organizations
- Carnegie Mellon University