Evidence integration for 3D object recognition: a connectionist framework

The authors present recent work on a vision system designed to recognize 3-D objects in a depth map. The system was originally capable of recognizing parametric surfaces of various types. The authors have added the ability to find parametric surface intersection curves and use these disparate types of information to index into an object database. From a depth map containing one or more objects, local surface and surface intersection curve estimates are determined. These are used in a series of layered and concurrent parameter-space transforms to extract the surface and intersection curves present in the image. Any one transform computes only a partial geometric description that forms the input to the next transform. The final transform is a mapping into an object database. An iterative refinement technique, motivated by work in connectionist systems, is used to integrate the evidence at each level. Fundamentally different types of evidence can be simultaneously extracted, be mutually supportive in intermediate levels of the recognition process, and cooperate to form a consistent interpretation of the image. Other features discussed include the modularity and consistency of the architecture