Sonar image quality assessment for an autonomous underwater vehicle

Abstract

Within the EU-funded project "ADVOCATE II" the participating partners are developing an advanced machine diagnosis system for autonomous systems, which is based on an integrated approach. The solution combines different "intelligent" modules to create the open software architecture for diagnosis and decision tasks. ATLAS ELEKTRONIK is going to integrate the ADVOCATE modules into an autonomous underwater vehicle (AUV), which must rely on an automatic obstacle avoidance system, based on sonar image processing. Beside typical electronic failures there is the possibility that the image quality is not sufficient for reliable obstacle recognition. The AUV needs to know this fact to react in an appropriate manner. To solve this sonar image assessment problem, a Bayesian belief module (BBN) has been developed. The BBN module is based on the AI technique known as probabilistic graphical models (PGMs). In particular, a time-sliced, object-oriented limited-memory influence diagram is used as the underlying PGM of the BBN module. The BBN module provides a diagnosis and suggests appropriate recovery actions on the sonar image assessment task