Computer vision for robust 3D aircraft recognition with fast library search

Abstract

A fast but accurate classification method is presented which takes the fundamental problems of aircraft recognition in 3D space into consideration. Elliptic Fourier descriptors are used for aircraft classification and pose determination from a two-dimensional image recorded at an arbitrary viewing angle. The computation model of the nearest neighbor classification rule (or, simply, NNR) is analysed to come up with proper necessary conditions used by our method for reducing the set of near neighbors of a test projection for a fast partial library search. On the other hand, our method is robust to noise by incorporating a perturbation analysis of feature variation into the classifier design. Both graphics-generated data and real data of five aircraft have been used to demonstrate the applicability of the new method. The results indicate that the method is almost as accurate as the NNR method, but is much faster. The method is also shown to outperform other existing classification methods including the decision-tree classification method.