Extended and Coordinated Targets: Tracker Design Consideration and Approaches

Abstract

Extended and coordinated target tracking involves advanced techniques for monitoring objects in dynamic environments, addressing challenges such as sensor resolution, the point target assumption and target behaviour. <br/> Sensor resolution is crucial, as higher resolution enables more precise detection and tracking, particularly in complex scenarios where targets may be closely spaced or vary in size and shape. The point target assumption, commonly used in traditional tracking, simplifies targets as single points. However, in extended target tracking, targets are modeled with more complex shapes or patterns, accounting for variations over time. <br/> Extended Target Tracking (ETT) becomes particularly important in scenarios where groups of targets move together, making it challenging to assign measurements to individual targets due to sensor resolution limitations and measurement noise. Even if the assignment problem is resolved, tracking each target individually within a group can be resource-intensive and impractical. To address this, Extended Target Tracking models the group as a single entity with shared kinematics, while simultaneously estimating the extent and shape of the group. This approach not only reduces computational demands but also enhances situational awareness. Examples of such scenarios include tracking fleets of aircraft or boats, formations of objects, swarms of drones, or groups of pedestrians. <br/> Also, ETT plays a vital role in Advanced Driver-Assistance Systems (ADAS), addressing challenges when a single object spans multiple sensor resolution cells. These systems require precise information about the positions, sizes, and shapes of surrounding objects to ensure accurate perception. ETT improves safety and performance, enabling vehicles to avoid collisions, execute safe manoeuvrers, and navigate complex environments, particularly when dealing with large or irregularly shaped objects in their surroundings. <br/> Group targets, such as fleets of aircraft, boats, or swarms of drones, present unique challenges when operating autonomously and in a coordinated manner. Advances in distributed and multi-agent systems have enabled these entities to collaborate at scales surpassing human coordination, achieving complex objectives. However, these innovations also complicate surveillance and defence efforts. Leveraging technologies like AI, communication, and computation, such systems achieve seamless coordination in dynamic environments, making them increasingly sophisticated and harder to counter. <br/> Coordinated Target Tracking (CTT) addresses the limitations of traditional surveillance systems by improving situational awareness, optimizing resource management, and enhancing threat intelligence. CTT provides operators with accurate risk assessments, enabling informed and effective decision-making. This approach is critical in applications such as surveillance and defence, where identifying coordination and evaluating threats are essential. <br/> In this dissertation, we will explore various approaches for simultaneously tracking both target shapes and kinematics by modeling target shapes as ellipses, particularly for LiDAR sensor measurements, and deriving a Gaussian mixture-based likelihood for improved accuracy. It introduces a novel Bayesian Gamma filter to estimate shape parameters, solving issues with negative estimates caused by Gaussian distributions. The approach extends to multivariate cases using the Wishart distribution, enabling robust and accurate shape parameter estimation. Unlike traditional Random Matrix (RM) methods, this work uses RM elements to represent shape extents, offering greater flexibility in modeling irregular shapes. These contributions enhance target tracking accuracy and reliability in dynamic environments. <br/> Additionally, we will discuss methodologies to enhance tracking systems when multiple agents or targets exhibit coordinated behaviour, distinguishing coordinated targets from extended targets and examining multiple forms of coordination beyond traditional group target scenarios. The use of accumulated state densities (ASDs) is introduced to identify coordination over time, improving the separability and clustering of group targets. A novel method leveraging velocity vector geometry infers the degree of correlation, demonstrating that targets moving toward a common point are statistically closer. This approach relaxes the typical assumption that targets must be closely spaced, enhancing the detection of coordinated behaviours in dynamic environments. These innovations significantly advance the understanding and tracking of multi-agent systems.