A COMPLETE AND EXTENDIBLE APPROACH TO VISUAL RECOGNITION

We present a framework for 3-D object recognition. An important aspect of this framework is its flexibility and extendibility, which is accomplished through a uniform, parallel, and modular recognition architecture. Concurrent and stacked parameter transforms reconstruct a variety of features from the input scene. These transforms are either based on data, data and reconstructed features, or combinations of reconstructed features. At each stage, constraint satisfaction networks collect and fuse the evidence obtained through the parameter transforms. This process ensures a globally consistent interpretation of the input scene and allows for the integration of diverse types of information. The final interpretation of the scene is a small consistent subset of the many initial hypotheses about partial features, primitive features, feature assemblies, and 3D objects computed by the various parameter transforms. This paper reports on a complete, integrated (and implemented) system that extracts planar surfaces, patches of quadrics of revolution, and planar intersection curves of these surfaces (lines and conic sections in three space) from a depth map viewing 3-D objects. The reconstructed primitive features are used to index into an object model database to form hypotheses about objects in the scene. Integration of the various modules is a significant aspect of this work. Experimental results detailing the recognition behavior of the system are presented.