Underwater localization with forward looking imaging sonar using artificial objects as references

This research focuses on the determination of the underwater position of remotely operated or autonomous underwater vehicles by simultaneously detecting and classifying artificial objects with forward-looking sonar simulation whose positions are known. As an alternative to costly and hard-to-reach underwater acoustic positioning systems, it presents an algebraic approach to determining the robot's position in a simulation environment by classifying the reflections detected in sonar using machine learning, employing distance-based multilateration, triangulation, and distance-based triangulation approaches with respect to these known objects. Additionally, in cases where sonar contact cannot be established, it accomplishes dead reckoning (DR) position estimation by combining inertial sensors on the robot. For achieving optimal accuracy in all calculations, a linear Kalman Filter is utilized to obtain the best results. The obtained values are compared with actual positions, yielding positive results in terms of the usability of a cost-effective localization approach.