Camera Calibration Explained: Mathematics, Concepts, and Practical Implementation

Camera calibration is one of the most important concepts in computer vision and robotics. Whenever a camera is used to measure objects, detect poses, reconstruct 3D scenes, or guide robots, calibration becomes essential. Without calibration, the camera image is simply a grid of pixels that has no reliable relationship to real-world measurements.

Calibration establishes the mathematical relationship between the 3D world and the 2D image captured by the camera. Once this relationship is known, the camera becomes a geometric sensor that can measure positions and orientations in space.

Calibrated using Charuco Marker

This article explains camera calibration in a way that students can understand and reproduce themselves. We will cover the mathematical model of camera projection, the calibration equations, and a practical step-by-step workflow that students can follow to perform calibration on their own systems.

Mathematical Model of Camera Projection

Intrinsic Camera Parameters

Intrinsic parameters describe the internal characteristics of the camera.

These parameters describe how the camera projects rays of light onto the image sensor.

Lens Distortion

Real camera lenses introduce distortion that bends straight lines.

Two main types occur.

Radial Distortion

Radial distortion causes lines to bend outward or inward.

It is modeled using:

x_corrected = x(1 + k1r² + k2r⁴ + k3r⁶)

y_corrected = y(1 + k1r² + k2r⁴ + k3r⁶)

Where

r² = x² + y²

Tangential Distortion

Tangential distortion occurs when the lens and sensor are slightly misaligned.

x_corrected = x + [2p1xy + p2(r² + 2x²)]

y_corrected = y + [p1(r² + 2y²) + 2p2xy]

During calibration, these distortion coefficients are estimated so that images can be corrected.

Extrinsic Camera Parameters

Homogeneous Transformation Matrix

This representation is commonly used in robotics and computer vision.

Final Calibration Result

Using the Calibration

Practical Tips for Students

Students performing calibration should remember the following guidelines:

Capture at least 20–30 images. Use different viewing angles. Ensure the calibration board covers different parts of the image. Avoid blurry images. Use accurate board dimensions.

Good calibration data results in low reprojection error and accurate geometric measurements.

Conclusion

Camera calibration connects the digital image captured by a camera to the real geometry of the physical world. By estimating intrinsic parameters, distortion coefficients, and camera pose, calibration allows computers to interpret images mathematically.

The calibration procedure involves capturing images of a known pattern, detecting feature points, solving the Perspective-n-Point problem, and computing transformations between coordinate frames. Advanced systems may also solve the hand-eye calibration equation to determine the relationship between cameras and mechanical systems.

Understanding this process enables students and engineers to build reliable computer vision systems capable of performing accurate measurement, localization, and robotic manipulation.