Date of Award


Document Type

Open Access Dissertation

Degree Name

Doctor of Philosophy in Electrical Engineering (PhD)

Administrative Home Department

Department of Electrical and Computer Engineering

Advisor 1

Timothy C. Havens

Committee Member 1

Michael C. Roggemann

Committee Member 2

Zhaohui Wang

Committee Member 3

Thomas Oommen


Sensor fusion is a process in which data from different sensors is combined to acquire an output that cannot be obtained from individual sensors. This dissertation first considers a 2D image level real world problem from rail industry and proposes a novel solution using sensor fusion, then proceeds further to the more complicated 3D problem of multi sensor fusion for UAV pose estimation.

One of the most important safety-related tasks in the rail industry is an early detection of defective rolling stock components. Railway wheels and wheel bearings are two components prone to damage due to their interactions with the brakes and railway track, which makes them a high priority when rail industry investigates improvements to current detection processes. The main contribution of this dissertation in this area is development of a computer vision method for automatically detecting the defective wheels that can potentially become a replacement for the current manual inspection procedure. The algorithm fuses images taken by wayside thermal and vision cameras and uses the outcome for the wheel defect detection. As a byproduct, the process will also include a method for detecting hot bearings from the same images. We evaluate our algorithm using simulated and real data images from UPRR in North America and it will be shown in this dissertation that using sensor fusion techniques the accuracy of the malfunction detection can be improved.

After the 2D application, the more complicated 3D application is addressed. Precise, robust and consistent localization is an important subject in many areas of science such as vision-based control, path planning, and SLAM. Each of different sensors employed to estimate the pose have their strengths and weaknesses. Sensor fusion is a known approach that combines the data measured by different sensors to achieve a more accurate or complete pose estimation and to cope with sensor outages. In this dissertation, a new approach to 3D pose estimation for a UAV in an unknown GPS-denied environment is presented. The proposed algorithm fuses the data from an IMU, a camera, and a 2D LiDAR to achieve accurate localization. Among the employed sensors, LiDAR has not received proper attention in the past; mostly because a 2D LiDAR can only provide pose estimation in its scanning plane and thus it cannot obtain full pose estimation in a 3D environment. A novel method is introduced in this research that enables us to employ a 2D LiDAR to improve the full 3D pose estimation accuracy acquired from an IMU and a camera. To the best of our knowledge 2D LiDAR has never been employed for 3D localization without a prior map and it is shown in this dissertation that our method can significantly improve the precision of the localization algorithm. The proposed approach is evaluated and justified by simulation and real world experiments.