Interleaving 3D Model Feature Prediction and Matching to Support Multi-SensorObject Recognition

摘要

This paper presents a system which performs on-line feature prediction for CADmodeled objects and tightly couples prediction with matching. For the ATR domain, detailed CAD models of objects are available in this application, as is both range and optical imagery. Matching begins with an initial hypothesis which is refined through an iterative generate-and-test procedure. Matching interleaves feature prediction and adjustment of model-to-sensor geometry until a locally optimal match is obtained. In addition, sensor-to-sensor geometry is also adjusted, allowing the algorithm to correct minor mis-registrations between range and optical imagery. While the resulting match is locally optimal in terms of the complete space of possible matches, it globally preserves the 3D constraints implied by sensor and object geometry. Results on real data are presented which demonstrate the algorithm correcting for up to 30 deg errors in initial orientation and 25m errors in initial translation.