Image understanding research at SRI International

The image understanding program at SRI International is a broad effort spanning the entire range of machine vision research. In this report we describe our recent progress in the core scientific domains of geometric scene modeling, delineation, and semantic labeling. We have addressed the problems of scene analysis in a number of application domains, primarily: modeling the earth's surface from aerial imaging sensors, ground-level scene modeling for robotic and weapon systems, and machine vision for industrial and medical purposes. In particular, we describe progress in developing interactive environments for scene analysis, context-based control of scene analysis algorithms, model based optimization, natural object recognition, passive ranging at night, geometric modeling based on integration of stereo and motion cues automatic site model construction from IFSAR data, wide-area terrain visualization, geometric modeling of micron-level structures for industrial assembly, and construction of 3-D anatomic models for surgical simulation.