If the image is known, I don’t know how the camera can calibrate the distortion through the camera.
1: Given only one image, it can be calibrated based on the relevant features in the image. Simply put, it uses the principle: line is straight.
2: Currently, the most commonly used method is to use a two-dimensional calibration plate to perform nonlinear optimization to minimize the reprojection error to achieve camera calibration. It is not calibrated based on the seemingly high-level training set.
3: The distortion parameters are only part of the parameters required by the calibration method, namely: two radial distortion coefficients and two tangential distortion coefficients. The purpose of eliminating distortion is to make the camera approximate the pinhole camera model as much as possible, so that the straight line will maintain its linearity when the camera images.
4: Common distortion correction algorithms are implemented based on this principle. Of course, there are two general situations.