Using imprecise environment information for guiding scene interpretation

We are concerned with translating knowledge about a real world environment into useful information regarding the content of images taken from this environment. In particular, knowledge about location and/or size of objects, expressed at different levels of imprecision, is translated into information on region allocation for these objects in an image when the position and camera angle are known (imprecisely as well). The results - objects which may have projection, areas of the image where they will be projected and information on their projection - can be used to guide an object recognition and scene interpretation system. Our approach is illustrated by real world - real image examples. (C) 1997 Elsevier Science B.V.