Robust checkerboard recognition for efficient nonplanar geometry registration in projector-camera systems

Projector-camera systems always need complicated geometry calibration to get a correct display result on nonplanar projection surface. Geometry registration of most calibration methods dealing with arbitrary surfaces is done by projecting a set of structure light patterns or by manually 3D modeling, which are both time-consuming. In this paper, we propose a robust checkerboard calibration pattern recognition method to help nonplanar surface geometry registration. By approximating the nonplanar surface to be composite of many planar quad patches, pixels mapping between the calibration camera and a projector can be got by projecting only one checkerboard calibration pattern recognized by our method. Compared with geometry registration with structure light or encoded points, which need project many images, our method can be more efficient. Our recognition method has two steps: corner detection and checkerboard pattern match. Checkerboard internal corners are defined as special conjunction points of four alternating dark and bright regions. A candidate corner's neighbor points within a rectangular or a circular window are treated as in different one-point-width layers. By processing the points layers in corner detection, we transform the 2D points distribution into 1D, which simplifies the regions amount counting and also improves the robustness against noises caused by deformation and complex illumination. After corner detection, the pre-known checkerboard grids rows and columns amounts are used to match and decide the right checkerboard corners from the results that have found. Regions boundary data produced during the corner detection also assist the matching process.