Härtill 5 uppsatser

Diss. (sammanfattning) Trondheim : Norges teknisk-naturvitenskapelige universitet, 2025

The reliable detection and tracking of maritime targets is a critical component in modern situational awareness systems, particularly for emerging applications in autonomous surface vessels and coastal surveillance. Traditional radar-based tracking systems are challenged by the prevalence of low signal-to-noise ratio (SNR) conditions caused by unwanted signal reflections from the sea surface as well as the frequent presence of low-observable targets, characterised by small radar cross sections (RCSs), such as kayaks, leisure boats, and floating obstacles. This thesis is concerned with these challenges, exploring robust probabilistic multi-target detection and tracking methodologies tailored to maritime environments with an unknown and dynamically varying number of low-observable targets. The primary objective of this research is to develop scalable multi-target tracking algorithms that can directly process raw radar measurements and address limitations under low-SNR conditions. At the core of this thesis is the development of a track-before-detect (TkBD) approach, termed integrated existence Poisson histogram-probabilistic multi-hypothesis tracking (IE-PHPMHT), which extends the underlying histogram probabilistic multi-hypothesis tracking (H-PMHT) framework through the integration of probabilistic target existence modelling utilising a Bernoulli target representation. This formulation enables flexible tracking of an unknown and time-varying number of targets, accommodating both target appearance and disappearance over time while retaining weak target information that would otherwise be suppressed by conventional detection thresholds. The thesis further employs an adaptive target birth strategy that enables data-driven track initiation. To complement this, a spatial gradient-based detector (SGBD) method is developed, in which improved detection capability is achieved by adapting image processing techniques to enhance the extraction of target features directly from raw radar intensity data, without relying on prior knowledge of target appearance. The core contributions are consolidated within a hybrid tracking architecture that combines the developed TkBD method with conventional detection-based tracking to balance computational efficiency and detection sensitivity. Through a full-scale implementation, the thesis demonstrates the real-world applicability of TkBD in complex maritime environments.