TOTAL BUILDING AREA EXTRACTION AND ORTHORECTIFICATION FROM HIGH RESOLUTION SATELLITE IMAGES WITH BUILDING DATABASE

초록

Research on orthorectification of satellite images is important for generating precise image maps, which are used in various studies such as change detection. Uncorrected relief displacements reduce positional accuracy of orthoimages. They are often caused by a height difference of man-made features. To remove them, we need height reference data for man-made objects such as digital surface model as well as digital terrain elevation models. It is inefficient to constantly update a digital surface model for man-made structures on frequently changing urban region. Unlike a digital surface model, a building database is updated during a construction process of individual buildings. Therefore, in this paper, orthorectification including building relief correction is conducted using a building database. Firstly, we detect building roofs and façades visible within high resolution images, which we refer to as total building area. In this process, we used the coordinates of building corners from the building database to calculate corresponding image points and define polygons for a total building area for each building. Next, we blank out the total building areas from original images and we generate orthoimages without building relief displacements. Next, we map the image patches of building roofs onto the orthoimages generated. Remaining blank areas in the orthoimages are from the occluded region due to height relief. These occluded areas are inpainted by neighbouring pixel values assuming. In this way, we generate orthorectified images with true building locations. We applied the proposed method to Kompsat-3A images with ground sampling distance of 55cm. We compared orthorectified rooftops from our method with a digital topographic map. In some buildings, errors occurred due to inaccurate height values within the building database used. However, when we adjusted the building heights with errors to true heights, errors were removed. Additionally, the inpainted areas appeared similar to building boundary regions in the original image. In this paper, we showed the feasibility of automated true orthorectification with correct building localization and occlusion inpainting. Based on this, it is possible to extract the building and rooftop area and use it to create a realistic image. Furthermore, our method could identify erroneous buildings within a building database and provide a mean to update them. © 2023 ACRS. All Rights Reserved.

키워드