Ego-Motion Estimation Using Doppler and Spatial Data in SONAR, RADAR, or Camera Images
Open Access
- Author:
- Monaco, Christopher D
- Graduate Program:
- Mechanical Engineering
- Degree:
- Doctor of Philosophy
- Document Type:
- Dissertation
- Date of Defense:
- September 23, 2019
- Committee Members:
- Sean N Brennan, Dissertation Advisor/Co-Advisor
Sean N Brennan, Committee Chair/Co-Chair
Hosam Kadry Fathy, Committee Member
Jacob Willem Langelaan, Outside Member
Russell Charles Burkhardt, Outside Member
Kurt A Hacker, Special Member
Karen Ann Thole, Program Head/Chair - Keywords:
- ego-motion estimation
doppler
sonar
radar
camera
autonomous
automotive
underwater
motion estimation
localization
perception
odometry
dead-reckoning
vehicle
spatial
doppler velocity log
navigation - Abstract:
- This dissertation focuses on the development of novel ego-motion estimation algorithms for automotive and underwater vehicles. Ego-motion estimation analyzes the perceived motion of the environment from onboard sensor(s) to estimate the vehicle's own motion within that environment, and thus ego-motion estimates are a critical asset for vehicle localization frameworks. Specifically, this dissertation focuses on utilizing RADAR or SONAR sensors with a particular emphasis on their Doppler measurements. This dissertation is comprised of several peer-reviewed research contributions. First, it presents background information and foundational research before presenting three novel ego-motion estimation algorithms: one reliant on measurements from a single RADAR, another reliant on monocular camera images and measurements from a single RADAR, and the last reliant on measurements from a 2D Forward Looking SONAR. All three algorithms rely on the same premise: decoupling ego-motion estimation to utilize different measurement domains for their respective strengths. Specifically, they utilize Doppler and spatial measurements for translational and rotational ego-motion estimation, respectively. This methodology has been shown to yield benefits in accuracy and computational cost. The presented novel algorithms have the potential to improve the localization accuracy of autonomous vehicles, particularly in denied low-visibility environments where an a priori map and/or landmarks are nonexistent.