2D – The algorithm locates points highlights work by the operator Difference of Gaussians and RANSAC invariant to changes in perspective and scale in order to make it possible to find the correspondence between the same regions in images of different dimension. By the set of points found is estimated a homography minimizing the reprojection error of least squares. The projective transformation of each frame according to the homography and the automatic correction of the brightness for the elimination of the solutions of continuity between the frames finally allows to obtain a mosaic image.
The software is developed in C ++ wrapper in C ++ / CLI, as a dynamic library for Windows.

3D – The estimate of the depth of each point of the work is obtained from a set of images taken by a camera on a robotic arm that moves horizontally with respect to the work.
Preliminarily calibration is performed inside of the camera with the method [1] with average accuracy of less than half a millimeter from low resolution images in order to obtain an estimate of the pose of the chamber with respect to the plane of traverse.
The motion of the room is used to select points of interest that matches dimensionally correct: notes arrays of internal and external calibration of the images captured during movement is calculated with precision of the fundamental matrix and then, for each point of interest detected, its ‘epipolar line’ which is the corresponding point. Finally it makes a triangulation dense to get away from a depth estimation.
The prototype is made of API Matlab, C ++ and OpenCV.
[1] Z. Zhang. A flexible new technique for camera calibration. IEEE Trans. Pattern Analysis and Machine Intelligence, 2000.