Real-Time Sonar Classification for Autonomous Underwater Vehicles.

Abstract : The Naval Postgraduate School autonomous underwater vehicle (AUV) Phoenix did not have any sonar classification capabilities and only a basic collision avoidance system. The Phoenix also did not have the capability of dynamically representing its environment for path planning purposes. This thesis creates a sonar module that handles real time object classification and enables collision avoidance at the Tactical level. The sonar module developed communicates directly with the available sonar and preprocesses raw data to a range - bearing data pair. The module then processes the range - bearing data using parametric regression to form line segments. A polyhedron building algorithm combines line segments to form objects and classifies them based on their attributes. When the Phoenix is transiting, the classifying algorithm detects collision threats and initiates collision avoidance procedures. The result of this thesis is a fully implemented sonar module on the Phoenix. This module was tested in a virtual world, test tank and in the first ever sea-water testing of the Phoenix. The sonar module has demonstrated real time sonar classification, run time collision avoidance and the ability to dynamically update the representation of the unknown environment. The sonar module is a forked process written in the 'C' language, functioning at the Tactical level. Source code and output from an actual Phoenix mission displaying the object classification of the sonar module are included.