Improved Multitarget Tracking in the Presence of Port–Starboard Measurement Ambiguity Using the Bayes Factor

We consider an application where an unmanned underwater vehicle (UUV) equipped with an acoustic sensor seeks to estimate the location of surface ships using relative angle measurements to the ships. The estimation problem is challenging due to ships occasionally appearing and disappearing from the sensor's field of view. On occasion, poor geometry between the sensor and the ships, and port–starboard ambiguity that is inherent in the sensor contribute to the challenges in the estimation problem. The latter challenge arises because the sensor cannot distinguish between sound sources on its port and starboard side (port–starboard ambiguity). Therefore, every measurement is associated with two possible sound sources that map each relative angle to bearing projections on the port and starboard side of the UUV. We approach the problem of identifying the origin of sound sources (relative angle measurements) that are most likely to be of actual ships using Bayesian hypothesis testing. We propose an assignment method that uses the Bayes factor as the criteria to recursively associate measurements to the target that is most likely to be an actual ship and not a false target. The effectiveness of our approach is evaluated by comparing the performance of three multitarget tracking algorithms with and without our method. Performance is evaluated offline using sea-trial data captured by a UUV developed by the Naval Research Laboratory and Bluefin Robotics.