An approach to classification of airborne laser scanning point cloud data in an urban environment

Abstract

Digital topographic data, including detailed maps required for urban planning, are still unavailable in many parts of the world. Airborne laser scanning (ALS) has the unique ability to provide geo-referenced three-dimensional data useful for the mapping of urban features. This article examines the performance of decision tree classifiers on two ALS data sets, collected in different seasons from different flying heights with different scanners using laser beams at different wavelengths – 1550 and 1064 nm – for the same study area. Classification was undertaken on the point clouds based on attributes derived from the triangulated irregular network (TIN) triangles attached to a point, as well as attributes of the individual points. Classification accuracies of 0.68 and 0.92 (kappa coefficient) could be achieved for the two data sets. Decision tree seems to be a classification method that is particularly suitable for geographic information system (GIS), as it can be converted to ‘if–then’ rules that can be implemented fully within a GIS environment. Grass and paved areas could be distinguished better using intensity from one data set than the other, which could be related to the wavelengths of the lasers, and need to be explored further.