Author

Semih Dinc

Date of Award

2016

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Computer Science

Committee Chair

Ramazan S. Aygun

Committee Member

Farbod Fahimi

Committee Member

Letha H. Etzkorn

Committee Member

Timothy S. Newman

Committee Member

Huaming Zhang

Subject(s)

Computer vision, Image processing, Robotics, Control theory

Abstract

Camera pose estimation is a critical stage for the vision based robotic control systems, since precise motion can be achieved by successful localization. Today, many robotic systems, such as unmanned vehicles, robotic arms, and other automation systems rely on vision systems. Additionally, emerging technologies, such as depth sensors and virtual/augmented reality headsets need accurate and reliable pose estimation algorithms. To serve this purpose, in this dissertation, a novel pose estimation algorithm called Mirage is proposed. Mirage estimates the camera pose in 3D Euclidean space by utilizing a desired camera pose and actual 2D view of a reference object. Mirage minimizes the 2D projection error between desired and actual pixel coordinates using analytical calculations in linear time. Using 3D Euclidean space and analytic calculations avoids undesirable Euclidean trajectories and complex computations. To evaluate the performance of Mirage, pose estimation experiments in simulated and real environments using noisy and noise-free data are also presented. Results of these experiments are compared with state-of-the-art techniques. An application of Mirage to the trajectory tracking problem is also reported. Results of trajectory tracking experiments using two simulated vehicle models and one real robotic system (two-wheeled non-holonomic ground vehicle) are included in that report. In all tested environments, Mirage generates fast and accurate results and outperforms other methods. This dissertation also introduces new solutions to two problems that need to be addressed before and after the pose estimation stage. First, it presents a new GPU based feature mapping method that generates reliable feature points for Mirage. Second, it introduces an error recovery algorithm that re-localizes the camera in case of pose calculation failure. Tests of these methodologies are also reported. The test results suggest these methodologies are promising

Share

COinS
 
 

To view the content in your browser, please download Adobe Reader or, alternately,
you may Download the file to your hard drive.

NOTE: The latest versions of Adobe Reader do not support viewing PDF files within Firefox on Mac OS and if you are using a modern (Intel) Mac, there is no official plugin for viewing PDF files within the browser window.