Camera calibration measures a camera's exact lens and geometry so a robot can turn pixels into real directions and distances — the unglamorous first step that every accurate vision system depends on.
Camera calibration figures out a camera's exact focal length, center, and lens distortion by photographing a known checkerboard from several angles. With those numbers, the robot can trust that a pixel really points in a specific direction.
Before a robot can trust anything its camera reports — how far, which way, how big — it has to know the camera's exact optics. Measuring them is camera calibration, and skipping it quietly corrupts every downstream result.
What it recovers
Calibration finds two sets of numbers:
Intrinsics — the camera's internal geometry: focal length, the optical center (principal point), and lens distortion (why straight lines bow near the edges of a wide lens).
Extrinsics — where the camera sits and points relative to the robot or world.
Together they form the map between a 3D point in the world and the 2D pixel it lands on.
How calibration is done
A known pattern seen from many angles gives enough constraints to solve for the lens parameters, which are then used to correct every future image.
Why it matters so much
Every metric vision task assumes calibrated pixels. Stereo vision depth is wrong if focal length is wrong. SLAM and visual odometry drift if distortion isn't removed. Visual servoing, 6-DOF pose estimation, and hand-eye coordination all inherit calibration error directly. A miscalibrated camera doesn't announce itself — it just makes everything slightly, consistently wrong.
In practice
The classic method (Zhang's) photographs a checkerboard from a dozen angles; the known square size gives the geometry needed to solve for the parameters. For a multi-camera or camera-plus-arm setup you also do hand-eye calibration to find where the camera sits relative to the gripper. Calibration isn't one-and-done: temperature, focus changes, and knocks shift the parameters, so precision systems re-check it.
Why it matters
Calibration is the foundation the entire vision stack stands on. Get it right once and everything above it can be trusted; get it wrong and no clever algorithm can fully recover.
