Characterisation of a complex maritime scene using Fourier space analysis to identify small craft

This paper describes the continuing development of a system, which is able to extract, identify and track maritime vessels. The system first interprets changes in the frequency domain of the image to build a characteristic frequency template that describes the structure of the current sea state. No a priori knowledge of any characteristic frequencies is assumed. The system is given various points within the sea and based on these points builds an average characteristic frequency response, effectively a template of the current sea state. The template is used to remove most of the sea from consecutive image frames, which leaves regions in the image mainly containing maritime vessels. Frame differencing of the frequency processed images is used to generate regions of interest, from which a set of object descriptors are calculated. A tracking algorithm resolves the frame to frame correspondences using a set of motion constraints. Regions that satisfy the constraints are highlighted and tracked. Regions that fail to satisfy the motion constraints are removed and no longer processed.