Urban Point Cloud Mining Based on Density Clustering 2 and MapReduce 3 4

Harith Aljumaily, Debra F. Laefer, and Dolores Cuadra 5 * Department Computer Science and Engineering 6 Carlos III University of Madrid 7 Av. Universidad 30 – 28911 – Madrid, Spain 8 {haljumai, dcuadra}@inf.uc3m.es 9 10 ** School of Civil, Structural and Environmental Engineering; 11 U3D Printing Hub & Earth Institute 12 University College Dublin 13 Newstead G25, Belfield, Dublin 4, Ireland 14 debra.laefer@ucd.ie 15 16 ABSTRACT: This paper proposes an approach to classify, localize, and extract 17 automatically urban objects such as buildings and the ground surface from a digital 18 surface model created from aerial laser scanning data. To achieve that, the approach 19 involves three steps: 1) dividing the original data into smaller, more manageable 20 pieces using a method based on MapReduce gridding for subspace partitioning; 2) 21 applying the DBSCAN algorithm to identify interesting subspaces depending on 22 point density; and 3) grouping of identified subspace to form potential objects. 23 Validation of the method was achieved using an architecturally dense and complex 24 portion of Dublin, Ireland. The best results were achieved with a 1 m sized clustering 25 cube, for which the number of classified clusters equaled that which was derived 26 manually and that amongst those there the following scores: correctness = 84.91%, 27 completeness = 84.39%, and quality = 84.65%. 28 29