Bayesian methods for Extended Object Tracking

This PhD thesis presents novel Bayesian methods for Extended Object Tracking (EOT), with a focus on developing robust shape models for improved tracking accuracy and computational efficiency. The work addresses key challenges in tracking maneuvering and occluded objects, object recognition, and simultaneous localization and mapping (SLAM). It introduces several original contributions: (1) a flexible prediction model for maneuvering targets (Λ:O model); (2) a lightweight elliptic tracker (L:OMEM); (3) a superelliptic shape estimator; and (4) a general-purpose non-parametric tracking and classification algorithm (FL:OREO). These contributions extend classical Bayesian filters to support both parametric and non-parametric shape estimation, enabling accurate tracking of complex, dynamic objects in real-time applications such as autonomous driving and surveillance.