Map-based Vehicle State Estimation Using A Spatiotemporal Preview Filter
Open Access
- Author:
- Leary, Robert D
- Graduate Program:
- Mechanical Engineering
- Degree:
- Doctor of Philosophy
- Document Type:
- Dissertation
- Date of Defense:
- June 03, 2019
- Committee Members:
- Sean N Brennan, Dissertation Advisor/Co-Advisor
Sean N Brennan, Committee Chair/Co-Chair
Matthew B Parkinson, Committee Member
Jacob Willem Langelaan, Committee Member
Karl Martin Reichard, Outside Member - Keywords:
- autonomous
vehicle
localization
perception
camera
pose estimation
mapping
lidar
state estimation - Abstract:
- The primary focus of this work is to develop a vehicle state estimation algorithm using a-priori knowledge of the environment. Specifically, this work focuses on the problem of achieving accurate localization of a vehicle within a map and on the road, using a map as a feedforward sensor to help estimate the location of the vehicle using image features. Presented here is a method for improving localization over standard GPS and inertial-based methods via map-based, monocular vision, state estimation algorithms. The measurements obtained from a camera pose estimation algorithm are fused with a dynamic vehicle model to improve vehicle state estimation in a real-time implementable algorithm. The presented methods, utilizing kinematic and dynamic modeling, allow for the calculation of the influence of specific three-dimensional road features when measuring a vehicle's pose. Additionally, the combined simulation and experimental implementation of these methods enabled comparative evaluations of the bounded region wherein the pose estimator can converge to the true vehicle pose under common road scenes using a map of the lane marker features. Finally, this work examines the use of a map-based Kalman filtering method using previewed road features and vehicle steering inputs, in coordination with the image-based pose estimation, to further improve the vehicle's state estimate.