A visual guidance calibration method for out-of-focus cameras based on iterative phase target

Calibration of out-of-focus cameras is crucial for some special 3d reconstruction applications. An essential factor in improving the calibration result is avoiding the solution's singularity which tends to pose the calibration plate at a large inclination. However, the depth of field (DOF) of cameras is limited, especially in the defocused scenes, and the blurred corners will significantly reduce the calibration accuracy. A vision-guided calibration method is proposed, including an optimal pose selection algorithm and an iterative phase-target technology. Compared with other phase-target methods, this method simultaneously optimizes the mathematical properties of the solution and the angular uncertainty of the corners. Compared with the Chessboard-target or Circle-target, this method has a higher calibration accuracy at three different distances that gradually change from focusing to defocusing. In addition, experiments also prove the effectiveness of the two submodules: Gaussian iteration and vision guidance. Finally, its code is available in the URL: https://github.com/SHU-FLYMAN/Visual-Guidence-Ca libration.