著者
Kosuke Takahashi Shohei Nobuhara
雑誌
研究報告コンピュータビジョンとイメージメディア(CVIM) (ISSN:21888701)
巻号頁・発行日
vol.2018-CVIM-212, no.39, pp.1-16, 2018-05-03

This paper addresses the use of mirror reflections for camera calibration. Camera calibration is an essential technique for analyzing the geometric and radiometric relationship between a 3D space and a 2D image. Most conventional camera calibration methods are based on a fundamental assumption: a camera can directly observe a reference object of known geometry. However, there are cases in which this assumption does not hold in practical scenarios. One approach to camera calibration in such cases is the use of a “mirror” as a supporting device. A mirror generates a virtual reference object that can be expressed using a small number of parameters. In addition, the 2D projection of the reflection object is equal to that of the known reference object from the virtual viewpoint. This paper utilizes these features and tackles two challenges of the geometric camera calibration; the first challenge is the intrinsic camera calibration when a known reference object is not available and the second challenge is the extrinsic camera calibration when the camera cannot directly observe a known reference object due to a physical constraint on the imaging system. The proposed algorithms introduce novel constraints, kaleidoscopic projection constraint and orthogonality constraint, which are hold with the mirror reflections for solving these problems. Evaluations with synthesized and real data demonstrates that the proposed algorithms can work properly and report the robustness of it in comparison with conventional methods.