A connectionist approach to primitive shape recognition in range data

The authors present the definition and construction of a vision system for recognizing 3-D objects whose surfaces are composed of planar patches and patches of quadrics of revolution. A significant percentage of man-made objects can be modeled using such surfaces. The authors concentrate on the subject of surface parameter estimation from noisy depth maps. Local surface parameter extraction is an important step toward 3-D object recognition. They propose a method that uses the curvature properties of polynomial approximations to the range data. The actual process of surface parameter extraction is framed as a set of hierarchical parameter-space transforms. Any one transform computes only a partial geometric description that forms the input to a next transform. An iterative refinement scheme is used to compute measures of confidence for geometric descriptions within the parameter spaces. This organization is motivated by connectionist visual recognition systems