Site-Model-Based Change Detection and Image Registration

摘要

The University of Maryland (with TASC as a subcontractor) is one of the group of institutions doing research on aerial image understanding in support of the RADIUS program. The emphasis of our research is on knowledge-based change detection (CD) using site models and the domain expertise of image analysts (IAs). Change detection involves classifying changes in the imagery as being due to site updates or activity, or as irrelevant changes due to illumination differences, seasonal variations, etc. The IA's expertise is crucial in identifying relevant changes, which depend on the site and the intelligence agenda. Our focus is on ways in which image understanding (IU) techniques can aid the IA in performing CD. We are designing a system that allows the IA to specify what are .to be considered as relevant changes, and to select appropriate IU algorithms for detecting these changes.rnBefore CD can be attempted, the acquired images have to be registered to the site model. We are developing efficient constrained search mechanisms for image-to-site model registration, using techniques based on non-monotonic reasoning (Assumption-based Truth Maintenance Systems (ATMSs) and their variants). We are also using such techniques to facilitate interactive IA guidance for CD and site model updating.