A surface matching approach to target recognition from forward-looking laser radar imagery

A modified spin-image surface matching system for out-of-plane rotation-invariant recognition in forward-looking laser radar (Ladar) range imagery has been described in the literature. In this paper, we experimentally address the relation between the training-set size used in the new method and recognition rate and present a method for selecting the training sample images. The modified system has attractive properties that include: rotated invariance, simplified training-set selection, predigestion from the 3-D range image to a set of 2-D images, tolerance for a significant amount of noise, no feature extraction. Each of these properties is investigated in the paper.