Camera Calibration

In computer vision applications, cameras project 3D world points onto 2D image planes. Traditional cameras possess intrinsic parameters including optical center, focal length, and lens distortion parameters that significantly influence the imaging process. Beyond these parameters, manufacturers typically provide technical specifications, though it's important to note that focal length changes during zoom operations, and other parameters may also adjust to minimize distortion effects. For depth information calculation and 3D scene reconstruction, precise camera positioning relative to reference points becomes essential. In stereo vision implementations, one camera can be designated as the reference point, requiring not only intrinsic parameters but also relative rotation and translation matrices between cameras. Standard calibration algorithms commonly employ chessboard patterns as reference objects. According to theoretical foundations, minimum calibration requires chessboard images from three distinct viewpoints, though increased image variety generally improves accuracy. However, using two or three orthogonal chessboards enables simultaneous acquisition of both intrinsic and extrinsic parameters in a single capture sequence.

This research presents a fully automated calibration technique using three orthogonal chessboards, applicable to both monocular and stereo camera systems. The implementation calculates extrinsic parameters based on intersection points of three orthogonal planes serving as reference coordinates. The algorithm workflow typically involves corner detection using functions like cv2.findChessboardCorners, followed by parameter optimization through least-squares minimization techniques.