Volumetric InSAS reconstruction using belief propagation

Interferometric synthetic aperture sonar (InSAS) is a technique for remotely estimating the height of the seafloor in a side scan configuration. Aperture synthesis provides a high along-track resolution and a pair of hydrophone arrays at different heights gives different views of the seafloor from which the seafloor height is estimated. Most height estimation algorithms use time delay estimation techniques (based on correlation) to compensate the footprint shift but rely on prior estimates of the seafloor height to correct the slight differences in signal stretching (wavenumber scaling) resulting from the projection of the sonar signals onto the seafloor at slightly different elevation angles. An alternative reconstruction method is to back-project the echo data into the volume to be reconstructed, similar to the volumetric techniques used in photometric computer vision for depth estimation from multiple cameras. With this approach it is simpler to use statistical inference techniques, such as belief propagation, and to include statistical models of prior information, such as scene continuity. In this paper the correlation and volumetric approaches are compared showing the advantage of incorporating scene continuity priors and how volumetric reconstruction implicitly handles the footprint shift and wavenumber scaling.