Integration of lidar data and GIS data for point cloud semantic enrichment at the point level

Commercial aerial laser scanning is generally delivered with point-by-point metadata for object identification, but current vendor-generated classification approaches (which rely exclusively on that data) generate high misclassification rates in urban areas. To overcome this problem and provide a fully scalable solution that harnesses distributed computing capabilities, this paper introduces a novel system, employing a MapReduce framework and existing GIS-based data, to provide more detailed and accurate classification. The approach goes beyond traditional gross-level classification (roads, buildings, trees, noise) by enriching the point cloud metadata with detailed semantic information about the object type. The approach was evaluated using two datasets of differing point density, separated by eight years for the same study area in Dublin, Ireland. As evaluated against manually classified data, classification quality ranged from 76% to 91% depending upon category and only 8% remained unclassified, as opposed to the commercial vendor’s classification quality which ranged from 43% to 78% with 82% left unclassified.