Extended Object Tracking: Introduction, Overview and Applications (1604.00970v3)

Abstract: This article provides an elaborate overview of current research in extended object tracking. We provide a clear definition of the extended object tracking problem and discuss its delimitation to other types of object tracking. Next, different aspects of extended object modelling are extensively discussed. Subsequently, we give a tutorial introduction to two basic and well used extended object tracking approaches - the random matrix approach and the Kalman filter-based approach for star-convex shapes. The next part treats the tracking of multiple extended objects and elaborates how the large number of feasible association hypotheses can be tackled using both Random Finite Set (RFS) and Non-RFS multi-object trackers. The article concludes with a summary of current applications, where four example applications involving camera, X-band radar, light detection and ranging (lidar), red-green-blue-depth (RGB-D) sensors are highlighted.

• The paper introduces a comprehensive approach that shifts from point-based models to tracking objects with extended geometric features.
• It details advanced methodologies such as random matrix models and star-convex shape estimation using Bayesian frameworks and nonlinear Kalman filters.
• The study demonstrates practical applications in surveillance, maritime, automotive, and robotics that optimize computational efficiency and accuracy.

Overview of "Extended Object Tracking: Introduction, Overview and Applications"

This paper provides an in-depth overview of extended object tracking (EOT), a vital field within multiple target tracking (MTT). The authors, Granström, Baum, and Reuter, focus on the transition from traditional point-based MTT models to the paradigm where each object may occupy multiple sensor resolution cells, thus necessitating the extended object tracking approach. The paper is structured to offer a comprehensive glossary of terms, rigorous methodologies, and practical applications, followed by compelling use-cases across diverse sensor types and object shapes.

Extended Object Tracking Basics

The transition to EOT is prompted by advancements in sensor technology, which allow multiple measurements per object, rendering the classical "small object" assumptions obsolete in certain applications. EOT can handle cases where not only the kinematic but also geometric properties of objects are essential for accurate tracking. Here, the paper explores modeling strategies, including:

• Shape Modeling: Various complexity levels in shape modeling are addressed, ranging from over-simplified geometric shapes to sophisticated arbitrary forms.
• Measurement Models: These models, fundamental to EOT, account for uncertainties and sensor noise, including Poisson point process (PPP) and binomial models, which estimate the spatial distribution of object measurements.
• Dynamic Models: Vital for predicting object motion, accommodating changes due to object turns and rotations.

Tracking Methodologies

The paper highlights two foundational approaches:

1. Random Matrix Models: These provide a Bayesian framework leveraging Gaussian inverse Wishart distributions for kinematics and extent. The paper details noise modeling improvements, showing enhanced fidelity to real-world scenarios by ameliorating biases in estimates caused by assuming negligible noise.
2. Star-Convex Shape and Gaussian Processes: Offering flexibility for arbitrary shapes, these use parametric or non-parametric modeling for complex shape estimation, applying Gaussian processes or random hypersurface models.

These approaches are empowered by nonlinear Kalman filters, capable of handling the complexities associated with EOT beyond traditional filtering techniques. Each methodology is critically analyzed for robustness and efficiency, with specific extended object tracking approaches demonstrating improved performance in simulation and empirical studies.

Multiple Extended Object Tracking (MEOT)

The authors discuss complexities associated with MEOT, where standard MTT challenges, such as estimation under occlusion or dense clutter, are compounded by additional computational burdens of extended object representation. The paper highlights reduced complexity approaches, focusing on distance partitioning and clustering techniques, which significantly optimize computational tractability without sacrificing accuracy.

Application Domains

The real-world applications of EOT explored include:

• Surveillance and Security: Tracking pedestrian groups using camera systems, effective in scenarios requiring crowd analytics or in arenas lacking individual object resolution.
• Maritime Domain Awareness: Using marine X-band radar to track ships, with accurate estimation of vessel dimensions aiding in classification and navigation safety.
• Automotive Safety and Autonomous Driving: Employing lidar sensors to assess vehicle kinematics and extents, crucial for advanced driver assistance systems and autonomous navigation.
• Complex Shape Estimation in Robotics: Utilization of RGB-D sensors for intricate object profiling, enabling environments where dynamic object interaction is crucial.

Conclusion and Future Horizons

The paper concludes by acknowledging several open research areas that require further investigation. These include the persistent challenge of computational load in real-time applications, performance evaluation metrics for generalized shape estimates, and the integration of diverse sensors for holistic environmental perception systems. By emphasizing these facets, the authors chart a path for future research which includes refining extended object models, addressing dynamic inter-object relationships, and enhancing the fidelity of real-time tracking systems.

Explored through theoretical frameworks and practical trials, the paper serves as a rich resource for researchers focusing on the progressive extension of MTT into realms where the comprehensive understanding of the object's geometry, in addition to its kinematic behavior, is necessary for application excellence.