Publication

Robust Auto-Calibration for Practical Scanning Setups from Epipolar and Trifocal Relations

Torben Fetzer, Gerd Reis, Didier Stricker

In: 2019 sixteenth IAPR International Conference on Machine Vision Applications (MVA). IAPR Conference on Machine Vision Applications (MVA-2019) May 27-30 Tokyo Japan IEEE 2019.

Abstract

The most important part of auto-calibration is the estimation of the fundamental matrices and the correction of the distortions caused by optical systems. From the fundamental matrices the state-of-the-art method of Lourakis[17] can determine the intrinsic calibration parameters. The prerequisite is that the fundamental matrices are very accurate, so that subsequent methods can converge. State-of-the-art methods minimize the epipolar error to approximate fundamental matrices. If more than two views are given, the trifocal error can theoretically also be used, but it is very noise sensitive and therefore less practical. In this paper, we propose a combination of both error types that leads to consistently improved fundamental matrices compared to the state of the art. The proposed method has been thoroughly evaluated on both synthetic and real data sets. Besides the increased probability that Lourakis' method converges, the resulting intrinsic and extrinsic parameters are of superior quality. The method is quasi parameter-free, easy to implement, and requires only a slightly increased computational effort.

EpiTriCalibration_Fetzer.pdf (pdf, 380 KB)

German Research Center for Artificial Intelligence
Deutsches Forschungszentrum für Künstliche Intelligenz