Dense 3D reconstruction using motion stereo for presentation of disastrous situations

Abstract

This paper proposes the method which creates less hole 3D model for the information of the disaster area. This method fills the holes of a point cloud model reconstructed with the feature based reconstruction method. To fill the holes, our method uses local and small dense 3D models reconstructed with the template matching (the area based method). And our method uses the superpixel to detect the holes of the point cloud model regardless of whether their contours are closed or not. Our method also uses the homography matrix to fit the view of the images which are captured from the different positions because template matching is not robust for the change of the view between images. This method makes easy to search the corresponding points because this method limits the search range for the corresponding points. And we confirmed an effect of the proposed method through an experiment.