Fusion of multiple side-scan sonar views

Abstract

A new set of techniques for the construction of large-scale side-scan mosaics is presented in this paper. The complete procedure operates in two main stages. First an accurate registration of the source side-scan images is performed by filtering navigation data, where an augmented state Rauch-Tung-Striebel (RTS) filter is used to smooth the stochastic map obtained from a Concurrent Mapping and Localization (CML) algorithm - which stores landmarks as extra states in a Kalman filter. Then a fusion algorithm based on Gabor wavelets is used to combine the registered side-scan images and assemble the final mosaic. The algorithm can be tuned to either maximize the information content or to minimize signal noise in the mosaic. The paper presents novel results created by fusing 14 views of a scene as observed by a side-scan sonar mounted on an Autonomous Underwater Vehicle (AUV). The resulting mosaics are tailored to preserve the texture information of the individual views, accentuate the returns from important objects in the scene, and cover a wider area than any one single view