Active contours for synthetic aperture sonar snippet registration

Abstract

In mine-hunting operations, there is currently a strong focus on using autonomous systems in order to remove personnel and ships from the minefield. These systems must perform their missions effectively, and the collected data should be exploited efficiently in order to maximize performance. Automatic target recognition (ATR) is a key enabling component of achieving this goal. Whilst immense progress has been made to design an optimal ATR, it still requires further improvement to perform at the level of a human operator in all environments. One way to increase automatic target recognition performance is to fuse multiple pieces of information on a given target, either, at data level (early integration), at feature level (intermediate integration) or at classification level (late integration). This paper details an algorithm for the registration and incoherent fusion of multiple views of synthetic aperture sonar snippets of targets at data level. The assumptions in this study are that data have been gathered on a target from opposing sides and targets have been detected with a basic detector.