Recognition of Polyhedral Objects from a Single Perspective Camera View

This work deals with the problem of recraizing polyhedral (or partially polyhedral) object from a single perspective camera view. Theproposed approach requires neither a priori knowledge of the environment nor time-consuming iterative procedures and is based on the solution of a linear system. The method belongs to the class of model based methods. The recognition is performed by using a distinctive local feature of an object. In this method it consists of a topological configuration of edges and vertices of the object. Models and image are modelled as graphs and a matching between feature of the model and a feature in the image is hypothesized by using graph theory. Then perspective transformation is found and the recognition procedure is applied to the hypothesized matching in order to verify the effective correspondence between the model object and the imaged object set of scaling coefficient is also found, one for each edge of the recognition pattern. In this way the method can deal with partially missed or overlapping edges.