Intermediate representation in model based recognition using straight line and ellipsoidal arc primitives

An intermediate representation suitable for the 2D recognition of 3D objects, from a single intensity image is proposed. Determination of the intermediate representation from CAD models and from intensity images is presented. This representation uses as primitives line segments and ellipsoidal arcs. A complete technique for fitting contour lines with these primitives has been developed. The method has been proved very accurate and robust to noise, thus it is suited for a variety of applications such as matching and recognition of objects in real-life images.