Multiple Extended Target Tracking in Maritime Environment Using Marine Radar Data
by Jaya Shradha Fowdur
Date of Examination:2022-07-13
Date of issue:2022-10-13
Advisor:Prof. Dr. Marcus Baum
Referee:Prof. Dr. Marcus Baum
Referee:Dr. Frank Heymann
Referee:Prof. Dr. Jesús García Herrero
Referee:Prof. Dr. Alexander Ecker
Referee:Prof. Dr. Ramin Yahyapour
Referee:Prof. Dr. Maciej Gucma
Files in this item
Name:MultipleExtendedTargetTrackinginMaritimeEnvi...pdf
Size:8.83Mb
Format:PDF
Abstract
English
Target tracking is today one of the main pillars supporting applications for Maritime Traffic Situation Assessment and Monitoring ( MTSAM ). It provides information on the targets (vessels) obtained from the sensors within an observation region of interest, allowing the users (from the on board captain to the port authorities) to get a more complete knowledge on the current traffic situation. Such knowledge could contribute significantly to detect and avert potential collisions, and could help traffic analysis by studying long-term trajectories of vessels. As sensor technologies have improved in terms of spatial resolution, measurements recorded from the Radio Detection and Ranging ( radar ) sensors –predominantly used for maritime navigation –occur in point clouds. This provides us with opportunities to estimate the kinematic properties of vessels as well as their extent (shape) information, a problem known as Multiple Extended Target Tracking (METT). The METT problem can be divided into two parts: Extended Target Tracking ( ETT) and Multiple Target Tracking ( MTT ). ETT considers extent estimation using basic shapes like the ellipse or more complex ones like the star-convex. Even for the basic shapes, the major challenge would be to find a trade-off between an accurate representation and the processing time, factoring in the measurement quality, for instance, in terms of noise level, spatial density of the point clouds and external influences such as weather conditions. MTT considers the problem of concurrently estimating the states of multiple vessels and the number of vessels itself. The approaches based on Data Association (DA ) to associate a measurement to its potential source often rely on a one-to-one constraint between them, and require to now cater for the association of a point cloud to its potential source efficiently. We propose two elliptical tracking approaches for the ETT problem, with particular focus on real-world marine radar data. The first one involves the opportunity to estimate the orientation (heading) of a vessel while keeping its dimensions fixed, targeting commonly encountered maritime-based situations where the heading is not aligned with the vessel’s course. The second one is another elliptical tracker which estimates the extended state of the vessel in a batch-fashion to help achieve the aforementioned trade-off. We then propose a custom DA -based MTT algorithm to process measurements that are priorly subject to a clustering approach so as to satisfy the one-to-one association constraint to handle the point cloud-nature of the measurements. The results have been evaluated using simulations and real data, and presented with comparisons (against state-of-the-art methods) and discussions. The final contribution is an approach for METT, which combines our batch elliptical tracker and our custom MTT tracker. The approach has been implemented on our demonstrator software that receives radar video streaming from harbours, in a multiple sensor-setting. To make the system autonomous and more robust, a track management scheme has also been integrated to maintain the tracks. We present random frame captures to illustrate the performance of our framework as a whole, for MTSAM.
Keywords: Extended Target Tracking; Multiple Extended Target Tracking; Marine Radar Data; Elliptical Modelling