Skip to Main Content
Geospatial imaging is a relatively new term which is increasingly becoming more important for both government and commercial sectors. Images taken at street level can be geo-coded using a camera equipped with a built-in GPS device. However, the location that GPS provides are prone to errors up to 10 meters. In this paper we propose an algorithm to find the accurate location of a street-level image taken with a fisheye camera within a satellite image. Our algorithm is based on straight line detection and matching using Hough transform and gradient information around the detected lines. The rotation parameter is obtained using the best corresponding lines. Then mutual information (MI) is used as the similarity measure along the best match lines to determine the translational parameters. Moreover, as the correction process is carried out for a consecutive series of images rather than an individual image, the final location of each image will be assessed to be consistent with its neighboring images.
Date of Conference: May 31 2010-June 2 2010