Interview with Daniel Köhntopp, Research Associate at Jacobs University and ATLAS ELEKTRONIK GmbH14-Feb-2017
You will be speaking at UDT 2017 on the topic of ‘Autonomous Mine Recognition using AUV and ATR'. Can you give us a brief insight into the areas you will be covering?
Naval mines can pose a huge threat to civil and military shipping alike. They are easy to deploy and highly efficient in relation to their inherent damage potential and cost of removal. Automating at least parts of the process by using unmanned vehicles can accelerate the removal significantly. Therefore, AUVs equipped with modern sonars are indispensable assets, in order to save time/costs and, most importantly, guarantee the safety of personnel. The session will be a brief introduction to the challenges and solutions of modern mine hunting applications. An overview of the used algorithms for the different sub-problems of recognising naval mines on sonar images is presented and example results are shown.
What can delegates expect to take away from your session?
The scenario envisioned for my talk is an Autonomous Underwater Vehicle (AUV) used in mine hunting applications. My talk will focus on the automated processing of the sonar image on the AUV right after it is generated. I will cover the multiple automated steps from the raw image to the recognition of a hazardous naval mine. In a first step the seafloor classification will be introduced as a means of estimating the complexity of the following recognition task. This recognition task starts with a rough but fast detection of objects of interest. Modern SAS systems like the ATLAS ELEKTRONIK VISION 1200 system are capable of surveying wide areas with high resolution. But most areas exhibit no mine-like objects and can thus be discarded, which is done in the detection phase. After the detection phase the sonar image is reduced to areas with mine candidates. The final discrimination between mine-like rocks and the different mine types is done based on the objects highlight and shadow visible on the sonar image. In order to get useful results, the noisy nature of sonar images has to be taken into consideration. Hence, a priori shape knowledge is used to mitigate the detrimental effects of the noise on the shape extraction process.
For more information, go to UDT 2017