Abstract

In the sequential approach to three-dimensional reconstruc- tion, adding prior knowledge about camera pose improves reconstruction accuracy. We add a smoothing penalty on the camera tra jectory. The smoothing parameter, usually fixed by trial and error, is automatically estimated using Cross-Validation. This technique is extremely expen- sive in its basic form. We derive Gauss-Newton Cross-Validation, which closely approximates Cross-Validation, while being much cheaper to com- pute. The method is substantiated by experimental results on synthetic and real data. They show that it improves accuracy and stability in the reconstruction process, preventing several failure cases.