Multi-Sensor Data Fusion in Localization and Tracking Applications with Performance Bounds

Abstract

Modern autonomous systems, such as unmanned aerial vehicles and self-driving cars, rely heavily on multi-sensor data fusion for robust localization and tracking. While integrating diverse sensing modalities enhances observability, it can also introduce redundancy, increasing computational complexity without proportional gains in performance. This underscores the need for principled frameworks that evaluate sensor information under uncertainty and resource constraints.

This thesis adopts a performance-based perspective using the Posterior Cramér-Rao Lower Bound (PCRLB) to assess estimation accuracy across diverse sensing modalities. By combining PCRLB analysis with tailored measurement models and fusion strategies, the work presents a unified methodology to guide sensor selection, configuration, and deployment in complex operational environments.

Three interrelated investigations are presented. First, in airborne angle-only tracking over uncertain terrain, a biased PCRLB formulation incorporates terrain elevation uncertainty and sensor bias. An estimation algorithm is developed to leverage opportunistic ground targets, supported by a joint filtering framework. This investigation is extended in a later chapter by introducing range-only measurements, terrain-informed CRLB formulations, and trajectory optimization strategies under terrain and bias uncertainty. Second, in GPS-denied scenarios, a decentralized cooperative localization framework is proposed using asynchronous inertial measurements. This system integrates pseudo-measurement fusion and rotation-aware uncertainty propagation, anchored by a tailored analytical PCRLB. Third, for radar-camera fusion in autonomous driving, novel measurement models and asynchronous CRLB formulations are introduced to support performance evaluation, fusion algorithm design, and sensor placement optimization.

Collectively, these contributions illustrate how performance bounds, when integrated into sensor data fusion workflows, can inform the design of efficient, reliable localization and tracking systems. The proposed methods are validated through comprehensive simulations and offer insights for deploying autonomous platforms in complex, real-world environments.