Autocalibration for Creating 2D/3D Displays on Complex Geometric Surfaces

Geometric calibration methods tend to rely on precalibrated cameras or real calibration targets. Here we explore the chal-lenge of achieving a sufficiently accurate projector system calibration for complex projection mapping tasks that can rely on neither precalibrated cameras nor calibration targets. Virtual scene models can take the place of a real calibration target for the purpose of providing scene information that is used to refine geometric estimates. Refinement of the display parameters aims to reduce the reprojection error, or the error between the theoretical coordinate (generated from the projection of the scene estimate coordinate by the estimated parameters) and the actual coordinate (obtained from pixel cor-respondences). For Christie installations, the final reprojection error of each pixel should be less than 1 pixel. Mystique was not always so capable. Christie’s research department, Christie Labs, has a history of creating solutions for new projection concepts. Part of that strategy has been a long partnership with the University of Waterloo, particularly the Vision and Image Processing (VIP) Lab in Systems Design Engineering. Graduate students study computer vision and machine learning—topics related to problems often at the heart of techniques—which can be used to bring new ideas, meth-ods, and solutions to Mystique and, by extension, to Christie’s customers. Years of investment in research and development, including partnership with the VIP Lab and others, have enabled Mystique to handle challenges beyond any other projec-tor-alignment solution. ID. © 2022, John Wiley and Sons Inc. All rights reserved.