Maritime Target Tracking with Exteroceptive Sensors and Target-Provided Information

Abstract

Autonomous and unmanned operations at sea are becoming increasingly common. However, as the operational goals increase in ambition, the existing methods show their weaknesses. The maritime domain can be complex and unpredictable, whereas the system demands in unmanned and autonomous operations are high. One of the most important pieces needed for maritime autonomy is situational awareness. Identifying targets and tracking their movements is a key component of situational awareness. This thesis concerns target tracking, with a focus on maritime applications. The main contribution of this thesis is a methodology for the utilization of targetprovided information in target tracking. In many applications, targets may transmit messages with valuable information regarding their identity, position, direction, and more. At sea, the Automatic identification system (AIS) protocol specifies the transmission of such messages and is mandatory for commercial ships and widely used by other vessels. Target-provided information has played a role in situational awareness since it became available, but often in cursory manners such as by plotting the position of the transmitting vessel on a map. This thesis combines the usefulness of target-provided information with advanced target tracking methods and explores how they, together with exteroceptive sensor measurements, can be an invaluable addition to the situational awareness of a vessel. Another contribution is the development, testing, and evaluation of the advanced trackers used to process the target-provided information together with exteroceptive measurements. The thesis presents an extension of the joint integrated probabilistic data association filter that includes support for multiple kinematic target models and modeling of target visibility. Both with and without the addition of target-provided measurements, the tracker is tested on maritime data and shown to be robust when faced with challenging problems. Furthermore, the thesis presents a variant of the Poisson multi-Bernoulli mixture filter with multiple kinematic models that includes target-provided measurements, with applications to both point target and extended object tracking. The results show that target-provided information such as position, speed, course, and vessel dimension generally improve the tracking results, and in certain situations, performance improvements can be considerable. The thesis also considers how to deploy such methods in real-life situations and presents a method for validating AIS message information. Validating the messages ensures their safe use when estimating the target states and enables their use in other parts of a large system. Furthermore, the trackers are tested in several fully autonomous collision avoidance scenarios, where the autonomous vessel has to safely avoid other vessels. The results show that the methods enable the vessel to make correct decisions and solve issues present with previous tracking methods used in similar situations. Lastly, some specific problems related to maritime target tracking are considered, namely time inaccuracies in the received measurements and false alarms from the wakes of targets. The presented solutions improve the tracking performance when the problems are encountered.