Incremental Large Scale 3D Reconstruction

In this paper, we describe our incremental large scale 3D reconstruction approach. The input for the algorithm is a set of overlapping sub-models, reconstructed from smaller sets of calibrated images. The output of the algorithm is a set of transformation matrices, that are used to bring sub-models to the common coordinate system. All sub-models transformed by their corresponding matrices form a single large model in a global coordinate system. The interesting property of the proposed approach is the possibility of in- cremental construction of the final composite model. The data collection, camera calibration, and computation 3D of sub-models can be performed independently by many users. Furthermore, adding new sub-models can improve overall reconstruction quality.