Camera Calibration

A camera turns three-dimensional rays into two-dimensional image coordinates. Calibration estimates parameters such as focal lengths, principal point, and lens-distortion coefficients. Many workflows observe a known planar target—such as a chessboard—from multiple positions, detect its corners, and solve for parameters that make projected points align with observed image points. OpenCV supports chessboard, ChArUco, and circle-grid patterns and describes calculating reprojection error from the difference between observed and projected points. Calibration quality depends on the input. Pattern dimensions must be correct, corner detections should be accurate, and images should cover varied positions, orientations, and regions of the frame. Many nearly identical views contribute less information than varied samples. Blur, glare, a cropped pattern, or a wrong grid size can produce poor parameters. Reprojection error is a useful diagnostic but can be low even when a target or model assumption is wrong; it is not an independent guarantee of metric accuracy. Calibrate the exact camera-lens configuration and image resolution used in the application. Validate on images not used in fitting, inspect residuals by view and image location, and test downstream measurements against known distances or geometry. If the lens, focus, zoom, resolution, or camera mounting changes, recheck the calibration. Store parameters with camera identifiers and version information. Calibration supports geometric tasks; it does not by itself recover depth from a single image or correct every motion, rolling-shutter, or environmental error.