Embedded multibeam sonar feature extraction for online AUV control

Abstract

For development of an intelligent unmanned autonomous underwater vehicle (AUV), sensor data needs to be processed online for navigation and mission planning. In this work, we suggest a complete workflow and a processing chain to retrieve multibeam sonar data for AUV control. Our approach is based on the well-known image processing library OpenCV which provides sophisticated image recognition algorithms. We implement a processing chain for feature extraction on multibeam sonar acoustic image which retrieves contours of objects and coordinate points of the contours. Coordinate points are discrete data which can be easily processed further with additional algorithms. E.g. size of objects can be determined with the coordinate points or an expert system can classify objects with help of the coordinate points. Our solution will be embedded in an online control of an AUV. We evaluate the performance of our feature extraction approach using pre-recorded sonar data.