Voxel Change: Big Data–Based Change Detection for Aerial Urban LiDAR of Unequal Densities

Abstract

The proposed voxel change (VC) algorithm provides accurate, scalable, and quantifiable change detection for urban aerial Light Detection and Ranging (LiDAR) scans. This VC algorithm uses MapReduce, a big data programming model, to map neighboring points into cubes. The algorithm converts each data set into a group of cubes, and classifies them into categories of building, ground, or vegetation. It then compares and quantifies changes in area or volume. Spatial discontinuity is overcome by clustering. Quality metrics are demonstrated by comparing a 1 km2 data set of Dublin, Ireland, using a 2007 scan with a point density of 225 points per square meter (pts/m2) and a 2015 scan with 335 pts/m2 (totaling more than 500 million points). By using only positional LiDAR information as the data input, the quality metric exceeded 90% across the full data set with respect to lost, new, and unchanged designations for vegetation, buildings, and ground areas, and regularly exceeded 98% for buildings. The technique successfully processes nonrectilinear features and robustly provides a quantification of change for both building expansion and vegetation at a 1 m3 level using dense, modern data sets.