3D building modeling based on 3D line grouping using centroid neural network

Abstract

Building reconstruction from aerial image data has been studied in this paper. 3D line segments generated by using stereo image analysis are usually fragmented, and it is very hard to reconstruct building rooftop from segmented 3D lines. Centroid neural network algorithm is employed to classify 3D lines into groups of lines. With this grouping technology, the grouped 3D lines are easily clustered into rooftop, and the 3D building model is reconstructed. The proposed approach is evaluated on the Avenches dataset of Ascona aerial images. This experimental results indicate that the grouped 3D lines can be efficiently used for the construction of 3D site models, and prove the efficiency of the proposed approach in dealing with the building reconstruction problem from complicated images.