Maritime Target Tracking with Varying Sensor Performance

Abstract

Several research and industrial projects have been started in recent years to drive the development of autonomous surface vehicles (ASVs). This thesis is one of the contributions to this emerging field, and the topic is on target tracking for maritime collision avoidance using a maritime radar. An ASV requires a full overview of the environment to plan and perform safe maneuvers. With exteroceptive sensors, such as a radar, ASVs are able to detect other vessels, also the ones that are not equipped with an automatic identification system (AIS) transponder. This is in particular relevant near harbors and urban areas, where there is a larger amount of vessels without AIS. At the core of this thesis is a target tracking system for ASVs based on the probabilistic data association filter (PDAF). This thesis makes several contributions on compensating for varying sensor performance. This is important for enabling smaller ASVs with low-cost sensors. In addition, it also makes contributions to sensor preprocessing, tracking system architecture and integration of target tracking with collision avoidance systems. The first contribution is on compensating for navigation system errors. An ASV is a moving platform, and uncertainty in the ownship state estimation system can be accounted for in the target tracking system. Second, the radar preprocessing and collision avoidance interfaces are described, as they are important components in practical implementations of target tracking for collision avoidance (COLAV). Third, the performance of track initiation methods are investigated. This includes a performance comparison using detection theoretic measures when the clutter and new target densities are unknown, as they often are in practical systems. Fourth, variations target detection probability is analysed, and three methods of estimating it are developed. Fifth, applications of the tracking system to maritime collision avoidance are presented. The research goals of the thesis is primarily motivated by real data obtained in the project. The results have been verified both in simulations and in full-scale experiments