Automated generation of urban Land Cover Maps and their enhancement and regularization

The aim of this article is to automatically generate and update Land Cover Map data for urban areas. The introduction examines the state of the art with a focus on enhancement of the classification results. The materials and methods of the processing are explained using a practical example. The applied classification method uses 18 features (normalized difference vegetation index, height above terrain and 16 attributes generated from four spectral bands) for each pixel of a digital true ortho-image. When training the classifier, only three small image patches per class were used. The enhancement of the classification results takes place in three steps. The first two steps create raster maps with smoothed outlines and generalized content. In the third step, straight, orthogonal, and parallel vectors are created for the outlines of buildings. The produced Land Cover Map of an urban area was checked for completeness and geometric accuracy. All buildings were detected, and the calculated standard deviations of building corner coordinates were σE = 1.0m and σN = 0.8m when the true ortho-image was used as reference. Possible improvements regarding source data, classification method, and enhancement are discussed. All processing can be done by open-source software, and a developed software package including documentation and examples can be downloaded from the Internet for own use. The results of this work can inspire both mapping organizations and amateurs to produce up-to-date thematic and topographic map data inexpensively and quickly.