Landmark extraction, matching and processing for automated image navigation of geostationary weather satellites

This paper addresses the issue on automated registration of images from weather satellites. Traditionally, weather satellite community has employed an approach called landmark detection for automated registration. A ground point or feature with known reference coordinates is defined as landmark. A landmark is matched against a weather satellite image. Based on match results estimated is the mapping function between images and a reference datum. This landmark detection approach has been suffered from the problem of mismatches. If match results contain errors, the accuracy of estimation deteriorates. To overcome this problem, we propose the use of a robust estimation technique called random sample consensus (RANSAC). Through intelligent strategy this robust estimator will distinguish inliers from outliers and establish the mapping function with inliers only. This estimator has been reported to work in land observation satellite applications as well as in many computer vision applications. We will show that the RANSAC can also work for our purpose. We tested our approach using a global coastline database and a GOES-9 image. A global coastline database was processed to generate 30 landmarks. They were matched against a GOES-9 image. Visible inspection revealed that the results contained 13 mismatches. With 30 match results the RANSAC was applied. It identified all 13 mismatches correctly. We can conclude that the RANSAC is able to select correct matches. For reliable automated registration, the RANSAC needs to be incorporated in the landmark detection process of weather satellite images.