With the increasing need for higher accuracy measurement in computer vision, the precision of camera calibration is a more important factor. The objective of stereo camera calibration is to estimate the intrinsic and extrinsic parameters of each camera. We presented a high-accurate technique to calibrate binocular stereo vision system having been mounted the locations and attitudes, which was realized by combining nonlinear optimization method with accurate calibration points. The calibration points with accurate coordinates, were formed by an infrared LED moved with three-dimensional coordinate measurement machine, which can ensure indeterminacy of measurement is 1/30000. By using bilinear interpolation square-gray weighted centroid location algorithm, the imaging centers of the calibration points can be accurately determined. The accuracy of the calibration is measured in terms of the accuracy in the reconstructing calibration points through triangulation, the mean distance between reconstructing point and given calibration point is 0.039mm. The technique can satisfy the goals of measurement and camera accurate calibration.

In camera calibration, surprisingly little attention has been paid to the whole calibration process, i.e., feature points extraction from image, the precision of the calibration pattern and the calibration method. They are of the same importance for the calibration result. To satisfy the special request of vision measurement system for camera calibration parameters, we present a valid camera calibration based on the maximum likelihood criterion using high precision virtual stereo calibration pattern, which is formed by moving an infrared light-emitting diode (IR LED) feature point with CMM on pre-defined paths. Radial distortion and decentering distortion are modled. By using bilinear interpolation square-gray weighted centroid location algorithm we can accurately determined the imaging centers of feature points image coordinate. During the calibration process, we adopt a linear parameter estimation and nonlinear rerinement based on the maximum likelihood criterion. The initial parameters values are computed linearly and the final values are obtained with nonlinear minimization based on the maximum likelihood criterion. With the accurate image coordinate and the space coordinate, this method can rapidly and validly converge, and experiment results show the maximum projection distance is 0.091mm, the mean projection distance is 0.0259mm, this calibration results can satisfy the request of the vision measurement system.

Camera calibration is a necessary step in computer vision. We presented a high-accurate technique for stereo camera calibration, which was realized by combining improved two-step method with virtual stereo calibration pattern, and considered radial distortion and tangential distortion. The most calibration parameters were estimated using a linear solution based on a pinhole camera model, we used the calibration points near the center of the image because of their little distortion for the solutions in the first step. In the second step, we took into account camera distortions and used all calibration points, the parameters estimated in the first step were improved iteratively through a nonlinear optimization. An infrared LED, which moved with three-dimensional coordinate measurement machine at certain distance, formed a virtual stereo pattern and given accurate calibration points. Using the calibration parameters computed by this technique to reconstruct calibration points, the accuracy is improved 0.11 mm compared with direct linear estimation.