Camera Clustering for Multi-Resolution 3-D Surface Reconstruction

In this paper we propose a framework for piecewise mesh-based 3D reconstruction from a set of calibrated images. Most of the existing approaches consider all available images at once. However, this is not tractable with very large sets of cameras. Therefore, we use subsets of images and evolve parts of the surface corresponding to those images. Our main contribution is an approach to partition the camera images, either semi-automatic, through clustering, or user guided, via a geometric modeling interface. The sub-parts of the surface corresponding to camera subsets are independently evolved at multiple mesh resolutions. This allows to handle large scenes and to increase the mesh resolution in surface parts containing high levels of detail at reduced memory and computational costs. We demonstrate the versatility of our approach on different data sets and with different camera layouts. Finally, comparing the piecewise and global reconstructions with groundtruth, we find no significant loss in the overall reconstruction quality.     «

In this paper we propose a framework for piecewise mesh-based 3D reconstruction from a set of calibrated images. Most of the existing approaches consider all available images at once. However, this is not tractable with very large sets of cameras. Therefore, we use subsets of images and evolve parts of the surface corresponding to those images. Our main contribution is an approach to partition the camera images, either semi-automatic, through clustering, or user guided, via a geometric modeling...     »