Image-based 3D Building Reconstruction Using A-KAZE Feature Extraction Algorithm

The development of 3D reconstruction from 2D building images enables cost-effective and accurate acquisition of spatial data. Feature extraction is a fundamental technique for 3D reconstruction method. However, buildings mostly consist of planar surfaces whose entities are feature-less. This study presents 3D building reconstruction using A-KAZE feature extraction algorithm. Because A-KAZE algorithm does not use Gaussian blurring like SIFT and SURF, A-KAZE algorithm has potential to extract correct visual features for feature matching and 3D reconstruction. The proposed method was tested on actual building scenes acquired from a high-resolution camera. The experimental results showed that the A-KAZE algorithm can detect the sufficient number of features with low computation time. It is expected that the proposed method can be implemented in comprehensive 3D reconstruction of civil infrastructures.