Interacting multiple model particle filters for side scan bathymetry

In this paper we propose a multiple sea floor model based approach to improve bathymetry estimation with tracking algorithms. Traditionally interferometry is used to estimate the phase difference of signals received by two sensors, implicitly the direction of arrival (DOA) of the wave impinging both sensors. In our approach, we employ a state space model to describe data collected by a multi-sensor side scan sonar, and the evolution of the underlying DOA angle. The challenge with space state models is choosing the right model, and detecting the switch between models. We propose the use of several models that describe different sea-floor patterns and merge them within the framework of the interacting multiple model (IMM). Since the sonar array processing problem is non-linear and non-Gaussian, we propose an IMM particle filter algorithm to provide robust tracking while not sacrificing performance. Also an interesting new application is the swath segmentation, which appears as a side result implied by calculating the different model probabilities.