Design and implementation of an algorithm for automatic 3D reconstruction of building models using genetic algorithm

Automatic extraction and reconstruction of objects from Light Detection and Ranging (LiDAR) data and images has been a topic of research for decades. In other words, laser scanner data are powerful data source for acquisition and updating of large scale topographic maps. With this information, topographic objects like buildings, trees and the relief can be determined. The goal of this research is to extract and delineate building ground plans from LiDAR data and reconstruction of buildings in 3D space. The focus of the research lies on the different possibilities to reconstruct the building models. In this paper, a reconstruction method based on genetic algorithms (GA) is presented by optimizing height and slopes of gable roof of building models. The proposed algorithm consists of three steps; initial building boundaries are detected in the first step. Then, in extraction step, in order to improve the accuracy of detection step, initial building contours are generalized and buildings are extracted. Finally and in reconstruction step, a GA-based method is used for reconstructing the building models. Also, the method has proved to be computationally efficient, and the reconstructed models have an acceptable accuracy. Examination of the results shows that the reconstructed buildings from complex study areas that uses the proposed method have root mean square error (RMSE) of 0.1 m.

► We present a novel approach to reconstruct complex 3D buildings based on genetic algorithm.
► The proposed method for building reconstruction uses genetic algorithm efficiently for optimization of a primitive polyhedral building model.
► This is a flexible model-based method which needs no predefined simple primitives.
► A comparison is made between the proposed GA-reconstruction method and the least squares method which proofs its performance.