IRIS-99-371

Research in Knowledge-Based Automatic Feature Extraction

R. Nevatia and A. Huertas

This report provides a description of USC's activities as part of the DARPA APGD (Automated Population of Geospatial Databases) project. Focus of this effort is on detection, delineation and description of 3-D buildings from aerial images. Buildings are objects of obvious importance in urban environments and accurate models of them are needed for a number of tasks such as mission planning, mission rehearsal, tactical training, damage assessment and change detection. An overview of a system for building detection and description using multiple views is given and its performance is evaluated. Range data contains direct 3-D information about a scene. However, the resolution of such data is typically lower than that of electro-optical panchromatic (PAN) images and the data may have many missing or highly erroneous elements. We propose to use range data to provide cues for more accurate reconstruction using the PAN images. It is shown that such cues can not only greatly improve the efficiency of the automatic system but also improve the quality of the results. Quantitative evaluations are given. An approach that allows a user to assist an automatic system in modeling buildings is described. The approach is designed to be efficient in user time and effort while preserving the quality of the models created. Currently our system is able to handle the rectangular buildings with flat roof or symmetric gabled roof. Models can be created by only one or two clicks in many cases. Efficient editing of automatically derived models is also possible. Results on selected windows of Fort Hood, TX and Fort Benning, GA, sites are presented. Current system is limited to rectilinear shaped buildings and assumes that ground height, camera models and illumination parameters are known.