Tracking road centerlines from high resolution remote sensing images by least squares correlation matching

This paper describes a semi-automatic algorithm for tracking road centerlines from satellite images at 1 m resolution. We assume that road centerlines are visible in the image and that among points on road centerlines similarity transformation holds. Previous approaches proposed for semi-automatic road extraction include energy minimization and template matching with global enforcement. In this paper we will show that least squares correlation matching alone can work for tracking road centerlines. Our algorithm works by defining a template around a user-given input point, which shall lie on a road centerline, and then by matching the template against the image along the orientation of the road under consideration. Once matching succeeds, new match proceeds by shifting a matched target window further along road orientation. By repeating the process above, we obtain a series of points, which lie on a road centerline successively. An Ikonos image over Seoul area was used for test. The algorithm could successfully extract road centerlines once valid input points were provided from a user. The contribution of this paper is that we proved template matching could offer wider applicability in feature extraction, and we designed a new template matching scheme that worked for feature extraction without global enforcements.