Experimental evaluation of segmentation algorithms for corner detection in sonar images

Abstract

The development of AUVs represents one of the latest great achievements of engineering and science for the exploration and monitoring of the underwater world. The use of these vehicles facilitates the collection of data and monitoring of these environments, allowing us to perform previously impossible operations. Autonomous navigation continues to present many challenges. The question of total autonomy is yet to be solved. Currently, AUVs are not able to navigate without some outside assistance for long periods. When underwater, due to the high attenuation suffered by the GPS and radio-frequency signals, the use of acoustic communications and sensors offers better results. Imaging sonars have been one of the most appealing technologies for acquiring information in underwater environments because they can yield acoustic images of the surroundings and have a set of particular characteristics that are well suited for obstacle detection and characterization tasks. Corners usually appear very distinct from the rest of the scene in sonar images, generally characterized by sharp intensities in a vertical direction. The detection of corners is particularly useful in human-structured environments such as tanks because the knowledge on their position provides a way to compute the vehicle position inside it. The combination of some basic operations typically used for image segmentation can be applied to the raw sonar image to detect and localize these spots. This dissertation proposes and evaluates with experimental data a set of image segmentation algorithms for corner detection in sonar scans. A detailed description of the necessary steps to accomplish this is provided as well as a critical analysis of the results following a few relevant metrics for autonomous navigation.