A modified topological descriptor for forward looking sonar images

The use of robots for underwater exploration has increased around the world in the last years. The automation of the underwater monitoring, inspection, and maintenance tasks often requires a mapping and localization system. One of the key issues of such system is to be able to recognize previously visited places through the sensory information. This paper proposes a modified method for description and recognition of acoustic images. The method builds a graph of the Gaussian probability density function that represents both the shape and the topological relation among the images. It was originally proposed by the first author and his co-authors in 2015. The modifications proposed herein include a new segmentation step for automatic estimation of parameters and a new graph comparison approach that does not need to make use of the orientation of the segmented regions of the images. The new approach was evaluated in a real experiment in a harbor area. It proved to be less dependent on the input parameters than the previous approach.