Angular Bundle Adjustment with Rotational Bias Estimation to Assist Multisensor Angle-Only Target Tracking
Muhammed Yeşilkaya’s Ph.D. thesis develops new methods for multisensor 3D angle-only target localization in the presence of sensor-orientation biases. The work formulates the problem as angular bundle adjustment, jointly estimating target positions and rotational sensor biases from synchronized angular measurements. A manifold-based formulation represents rotations directly on SO(3) and applies Riemannian trust-region optimization, improving finite-iteration behavior especially in weak and long-range sensor-target geometries.
The thesis also introduces a complementary hybrid ML/MAP formulation on Euler angles, where prior statistical information on sensor biases is combined with the angular measurements. This approach improves localization accuracy when reliable calibration or navigation uncertainty information is available. Together, the two methods provide a geometry-aware framework for robust passive multisensor localization and form a basis for future extensions to time-varying biases, synchronization errors, and recursive target tracking.