Multiple Extended Object Tracking in Confined Waterways

Abstract

There is a growing interest in the development of autonomous systems for maritime vessels. One reason for this interest is the potential of maritime autonomy to revitalize the use of inland waterways as a sustainable and efficient means of transport. However, achieving autonomy in these settings is fraught with unique challenges. Confined waterways, such as harbors and canals, require sophisticated situational awareness systems to ensure safe navigation. One challenge is the close proximity to other vessels, which makes the representation of another vessel as a single point insufficient. Instead, extended object tracking methods are required, where the vessel’s extent, meaning its size and shape, is estimated. Another challenge is that the bounds of the waterway and other static elements in the waterway, such as quays and navigational marks, also need to be considered. This thesis explores the application of multiple extended object tracking to address the challenge of situational awareness in confined waterways. It uses an existing state-ofthe- art multi-object filter, the extended object Poisson multi-Bernoulli mixture filter, and combines it with a detailed extended object model, the Gaussian process extent model. This model is particularly well suited for high-resolution sensors such as LiDARs since it models the extent with a parametrization of the contour. The thesis also presents enhancements to this filter. One such enhancement is the incorporation of information from the Automatic Identification System (AIS) and a specific method to fuse this information with data from other sensors within the framework of the filter. Another enhancement revolves around the use of negative information, the absence of information, as a source of information. This is used to improve the state and extent estimation. This concept is further extended to model occlusion of different objects within the filter. To address the challenge of distinguishing vessels from static elements of the environment like waterway boundaries and quays, the thesis combines the developed tracking framework with a mapping approach. This fusion enables the simultaneous mapping of static environments and tracking of dynamic vessels as extended objects. The result is a situational awareness system tailored to the demands of confined waterways. The thesis features extensive use of experimental data which has been used to validate the proposed methods. This involves both small pleasure craft as well as larger inland barges, which are both able to be tracked using the same fundamental framework that is presented in the thesis.