超解像を利用すればStructure from Motionによる３次元復元モデルの精度は向上するか？

丹　康弘, 丹　佑之

Abstruct

The structure from motion (SfM) technique with forward-looking photogrammetry is one of the most widely used 3-D modeling techniques for obtaining a digital surface model of an object and landscape. The accuracy of point clouds depends on the image resolution of a camera. It is difficult to obtain a detailed 3-D model with a cheap camera deployed on a small unmanned aerial vehicle. We present a novel approach to enhance the reconstruction precision of a 3-D model. The proposed method applies SfM technique to super resolution images employing a Gaussian mixture model. Experimental results show that the digital surface model generated using the proposed method is within the acceptable range of GNSS-TS surveying, and our method surely provides improved reconstruction accuracy for a 3-D model for a large region.