Date of Completion

5-20-2019

Embargo Period

5-20-2020

Keywords

Sensor Fusion, Target Tracking, Heterogeneous System

Major Advisor

Yaakov Bar-Shalom

Associate Advisor

Peter Willett

Associate Advisor

Krishna R. Pattipati

Field of Study

Electrical Engineering

Degree

Doctor of Philosophy

Open Access

Campus Access

Abstract

Target tracking, motion estimation as well as localization problems are widely studied due to their importance in defense systems and self-driving vehicle systems. There are two related research areas covered in this thesis.

(i) Trajectory estimation of a thrusting/ballistic target using a single fixed passive sensor. Although a sensor network can provide higher accuracy, using one single sensor is worth considering because it is economically efficient. A Maximum-Likelihood (ML) estimator is developed for the target motion parameter estimation. The statistical efficiency of the estimator is evaluated using the Normalized Estimation Error Square (NEES). A discussion of the observability is provided via the uniqueness of the target state vector for a certain sequence of 2-dimensional angle-only measurements from a single fixed passive sensor. Starting with a polynomial motion, the results are extended to nonlinear thrusting/ballistic motion.

(ii) Sensor Fusion for heterogeneous and asynchronous systems. In a heterogeneous system, the state models used by the local sensors are in different state spaces with different dimensions. One reason for using distinct system models in the local trackers is the different sensor characteristics — active vs. passive. Considering, e.g., automotive environment perception, heterogeneous tracks are inevitable due to different coordinate systems, which are related by a nonlinear transformation with no inverse. The cross-covariance between the process noises used by the heterogeneous local tracks (LTs) is derived. Two fusion techniques are introduced in the thesis for a heterogeneous and asynchronous system: Track-to-Track Fusion (T2TF) with cross-covariance and Information Matrix Fusion (IMF), which does not require the cross-covariance between the estimation errors.

Available for download on Wednesday, May 20, 2020

Share

COinS