Fusion Algorithm Creation for Advanced Detection via Evolution (FACADE)

Sensor technology has recently advanced to the point where it is has become quite feasible to use multiple sensor modalities on the same platform. However, methods for intelligently fusing the data from the disparate sensors into a more descriptive form have proven illusive. Fusion of features not requiring co-registration of the image data is obviously beneficial; however, it is still an open problem as to which features are useful for ATR tasks and also sufficiently stable to fuse across sensor modalities. We present a set of candidate features for electro optic, infrared, and LADAR images that are then fused. These features include: spin images computed at 3D feature points, vertical surface features, Gabor functions, and scale-space features. We outline a method by which an evolutionary algorithm takes ATR training data and its associated ground truth to discover an optimal fusion for a set of candidate features, given the desires set forth by the decision-maker (such as desired point on ROC curve). We discuss an evaluation plan for determining which candidate features are most suitable for fusing. Additionally, we explore an optional scale-space based image registration allowing pixel-to-pixel fusion to also be performed.