Parallel Algorithm for Building Extraction from LiDAR Data

The data presented by Light Detection and Ranging (LiDAR) systems are in a dense and accurate three dimensional pattern without point classification, such as trees, roads, and buildings. Extraction of boundary points is essential for recognizing buildings. However, it is complicated to process the LiDAR data due to its irregularity and a large number of collected data points. In order to find boundary points in a quick and accurate way from a huge number of data points, a parallel algorithm for building extraction is proposed. In this paper, every processor reads the LiDAR data points and builds a quadtree respectively. Thus, a quadtree is built and shared through a network file system. Later, a breadth first search (BFS) is applied on every processor. The position of a node in the quadtree, in which has the number of children smaller than a predefined grain size, is stored in an array called as BFSArray. Process farming paradigm has been applied to process each nodes using MPI. In our primary experiment, results show significant speedup for multiple processors compared to a sequential algorithm.