Automatic Camera Calibration and Model Reconstruction Using Multi-View Stereo

Abstract

In this paper, we propose a method for automatic camera calibration and model reconstruction from image sequences. Our camera calibration method is based on the recent works of structure from motion that recover camera motions via a set of corresponding feature points extracted from images. The refinement of camera parameters is achieved by exploiting the concept of silhouette coherence. Eventually, a modified volumetric graph-cut algorithm is used for model reconstruction from the set of calibrated images. We use an occlusion robust photo-consistency metric, which does not require geometric heuristics about the object, to seek the optimal surface that captures the concavities. At the same time, a set of surface points identified from silhouettes are enforced as hard constraints to preserve protrusions. The camera calibration results of the proposed method are comparable to that of manual calibration even on real data, and the results of model reconstruction are also satisfactory. Since our method does not require images being calibrated in advance, it is suitable for digital archives and virtual exhibition of museum artifacts.