Multi-Hypothesis Automated Wireframe Generation

"Accurate targeting using standoff missiles, such as JASSM, requires automatic target correlation algorithms using significant target features. Wireframe models that describe the significant edges and corners of fixed targets are well suited for this task.Great advances in wireframe construction from stereoscopic imagery have been achieved through the last twenty years of research in computer vision, but significant limitations still exist. Current model construction procedures are human-intensive and timeconsuming. Existing automated systems create too many spurious objects and incorrect hypotheses that lead to incorrect scene surface topologies, caused by physical phenomena such as surface reflectance, ambient lighting, shadows and occlusions generatedby complex scenes. We will make automated wireframe model algorithms truly useful for mission planning platforms by rigorously applying physical and probabilistic models to a wider set of image features. First, we will use a rigorous mathematical treatment of uncertaintiesto better characterize the data and avoid sensitive heuristics. Second, we will develop new probabilistic, phenomenologically motivated cueing algorithms to distinguish false alarms from targets. Finally, we will enhance algorithms to consider multipletarget hypotheses, with a model-based predict and match algorithm to robustly choose among competing hypotheses using predictions of edge, area, and texture