3-D model based robot vision by matching scene description with the object model from a CAD modeller

Abstract An algorithm for Robot Vision is proposed which matches scene description to object models in order to determine the location and the orientation of the keysurface of an object lying on the worktable. The object is modelled using a three dimensional cad modeller. The geometrical data of the object, accessed from the cad/cam interface, is then used to calculate all the stable positions of the object. For each of these stable positions, the characteristic parameters derived from the planview boundary is stored in a database. A matching technique is then employed to compared the 2-D image of the planview of the real object with these parameters to determine the stable position of the object. A set of visibility rules, generated for the specific object, is then executed sequentially, in which the special identification of some operation-oriented keysurface of the object is compared with the 2-D image of the elevation and endview to ascertain the spatial orientation of the keysurface.