Robust Vehicle Localization using GPS, In-Vehicle Camera, Magnetic Guidance and Kalman Filtering

Open Access
Mangus, Anthony Joseph
Area of Honors:
Mechanical Engineering
Bachelor of Science
Document Type:
Thesis Supervisors:
  • Sean N Brennan, Thesis Supervisor
  • Sean N Brennan, Honors Advisor
  • Henry Joseph Sommer Iii, Faculty Reader
  • Kalman Filtering
  • GPS
  • Vehicle Estimation
  • Localization
This research focuses on reducing the effect of sensor faults and noise on the lateral vehicle estimation problem. The lateral estimation algorithm aims to localize the vehicle within the confines of a lane, given known and unknown faults. In order to test these algorithms, the Pennsylvania State University Rolling Roadway Simulator (PURRS) was reconfigured and simplified; however, to reduce complexity of testing, the treadmill belt was not used and the vehicle was moved by hand. In addition, a new vehicle was designed and built to provide a more rugged and utilitarian vehicle for use on the PURRS. In this work, these hardware changes are discussed, as well as the development of a Magnetic Guidance Calibration Stand (MGCS). These hardware systems are then used to develop fault reduction algorithms for use with a vehicle equipped with two magnetic sensors, an in-vehicle camera, and simulated GPS sensor. Two algorithms are tested: one to reduce the effect of an unknown fault, such as a sensor failure, and the other for known faults, such as a known change in environment that increases measurement noise. These algorithms were tested off-line using data collected using physical hardware on the PURRS. These algorithms are shown to reduce the effect of a fault on the estimation of the vehicle's position.