3-D object matching in the Hough space

An object-model matching approach is presented in this paper that considers the object in the scene being the result of a transformation (rotation and translation) from its model, both of which are represented by the edges and vertices of the surfaces. Features are extracted from range images so that information of edges and vertices are in 3-D space. To determine the rotation, each edge in the model is matched against every edge in the scene with the help of an arbitrarily pre-selected reference vector. Each such match determines a rotation matrix. Then, translation parameters are computed using the vertices of the edges using this rotation matrix. A 3-dimensional Hough space representing the three translation parameters is used for the translations from all possible rotations. The translation at the highest peak in the Hough space and its corresponding rotation are determined as the model-to-object transformation. The experiments indicate that the procedure is accurate and rather efficient