Progress toward a simple, parallel vision machine
Any intelligent agent, be it robot or human, needs to continually update its knowledge about the surrounding environment, and images offer a rich source of relevant information. Unfortunately, however, in normal two-dimensional images, the effects of viewing geometry, illumination, surface reflectance, and object shape are confounded together in a way that makes it very difficult to extract interesting information about the surrounding scene.
Access to the full text of this article is restricted. In order to view this article please log in.