zql@aipna.ed.ac.uk (Ziqung Li) (07/13/89)
A computer vision system based on neural networks is being built. It is to be called VisionNets. The VisionNets is built for 3-D object recognition and location from range image. It is expected to deal with free-form objects in a robot environment. It consists of 3 levels of neural nets based on the Hopfield models: 1. Low level net: from a range image to surface curvature images (work finished, tested using real range data, well) 2. Intermediate level net: from the suface curvature images to invariant surface descriptions in attributed grapghs (work finished, tested using real range data, well) 3. High level net: from the invariant surface descriptions to object classification and location (programs go through, tested in the recognition phase, work; but not the location phase owing to the lack of a 3-D object model) I am looking for geometric models to test the VisionNets as well as the 3rd level of the Nets. Any information about available existing 3-D object models, such as CAD models of free-form objects or any other explicit models, would be greatly appreciated. Thanks in advance. Z. Li zql@aipna.ed.ac.uk zql%uk.ac.edinburgh.aipna@ucl.cs.nss