Clustering and feature extraction in a 3D real-time echo management framework

Abstract

The on-board perception process is a fundamental part of the autonomous underwater vehicle (AUV) expert system. The possibility of using the sonar to build a real-time mental image of the environment is specially interesting in this perspective. This paper describes the research using a real-time framework for autonomous sonar based 3D perception. A spherically organised map is used to continuously keep track of the spatial relationships between echoes. This local map is intended to support several of the AUV processes with sonar based information, for example collision avoidance at the low level and feature extraction for object modelling and identification at a higher level. Experiments in the Trondheim fjord has resulted in promising cluster maps where the majority of 3D sonar returns corresponds to the target objects. An approach for real-time feature extraction and object modelling within this framework is presented. It is specially illustrated that when this sonar image is made at the time of echo acquisition, real-time echo clustering and surface modelling becomes feasible using inexpensive hardware.