Camera Calibration

Estimating a camera's intrinsics, distortion coefficients, and extrinsics so pixels and 3D coordinates can convert into each other.

Camera calibration is the process of estimating a camera's imaging parameters: its intrinsics (focal length, and the principal point where the optical axis meets the image), its lens distortion coefficients (radial and tangential), and its extrinsics (rotation and translation relative to some reference frame). The most widely used method is Zhang Zhengyou's, published in IEEE TPAMI in 2000: it shoots a planar calibration board from at least two different poses, computes a closed-form solution first, and then refines it with maximum-likelihood optimization. Calibration quality is judged by reprojection error — reprojecting 3D points back onto the image using the estimated parameters and comparing the pixel distance to the points actually detected. Without accurate intrinsics and extrinsics, a robot can't turn a depth map into a point cloud or map vision results into arm coordinates; depth cameras usually ship with factory-calibrated intrinsics, but the relationship between the camera and the arm still has to be calibrated separately.

ExampleProcessing twenty-some checkerboard photos with OpenCV's calibrateCamera gives fx, fy (focal length in pixels), the principal point cx, cy, and a set of distortion coefficients, along with an average reprojection error — the smaller that number, the better the calibration.

Also called

Intrinsic Calibration, Zhang's Calibration Method

Related

Camera Intrinsics · Camera Extrinsics · Lens Distortion · Calibration Board · Hand-Eye Calibration · Pinhole Camera Model

Sources

Zhang, A Flexible New Technique for Camera Calibration (IEEE TPAMI 2000)
MATLAB: What Is Camera Calibration?