Model-Free Multi-Camera Calibration By Graph Pyramid Using 360° Pattern

Camera calibration should be easy to perform on the one hand and provide robust results on the other hand. These requirements motivate the proposed, "unusual" approach for multi-camera calibration. For camera calibration, common methods assume a geometrical model to correct distortion. Here, however, the proposed method introduses an approach that does not make any assumptions about the projection model of the lens and its parameters and thus allows nearly arbitrary distortions. This calibration procedure employs a special 360º pattern including different markers and fractal shapes in different resolutions which applicable in particular to multi-camera setups. For such multi-camera setups, not only a calibration is needed for each camera, but the individual images have to be aligned to fit together in a final image mosaic. Image alignment is usually considered a separate problem from distortion correction and thus handled separately. The proposed approach, however, combines distortion correction and image alignment in a single procedure. This new approach employs graph pyramids to perform this registration problem and entails several advantages. First, it does in one procedure both distortion correction and image alignment using a specific 360º pattern. Second, it includes highly parallelized processes meants that individual images are prepared by the cameras themselves and the computational cost at a central processor is reduced. Finally, there is no need to detect key points and solve the correspondence problem in order to do image stitching.