This work presents a camara calibration procedure to be applied in a stereo vision system used in visual inspection activities involving the three-dimensiona; reconstruction of a scene. The presented procedure encompasses a robust method developed for the solution of non-linear least-squares problems encountered, in order to obtain global solutions to these problems and achieve the smallest error in the parameters to be extracted.

    For the development of this robust method several optimization techniques were used, from a theorical analysis of the problems to be solved. The initial estimation was provided using a linear method derived from the projective geometry theory. A solution for the problem of finding the best matching between corresponding points in stereo image pairs is included. Distortion effects introduced by the camera lenses, as well as a method to correct the errors introduced, are described.

    Tests with real images were performed to verify the robustness of the proposed method, inclusively under noise effects. The errors obtained are less than 3% in the mean, which are considered satisfactory for the application in view.