Introducing An Efficient Set of High Spatial Resolution Images of Urban Areas to Evaluate Building Detection Algorithms

Abstract

The present work aims to introduce an efficient set of high spatial resolution (HSR) images in order to more fairly evaluate building detection algorithms. The introduced images are chosen from two recent HSR sensors (QuickBird and GeoEye-1) and based on several challenges of urban areas encountered in building detection such as diversity in building density, building dissociation, building shape, building size, building alignment, building roof color, building height and imaging angle. In order to practically examine the proposed dataset, three building detection algorithms with different strategies are employed. The results imply the proposed dataset can be helpful to more fairly evaluate each algorithm, so that it indicates where the algorithm can be efficient and successful and where may be encountered with the problems in detecting buildings in urban areas.