Date of Award
Master of Science
Department of Electrical and Computer Engineering
John F. Raquet, PhD.
There are a large number of commercially available quad-rotor helicopters available from various manufacturers. All of these systems rely on a low cost MEMS based inertial measurement system for stabilization and navigation. These low cost inertial systems are all subject to rapid error growth in their attitude and position estimates unless bounded by external measurements. This thesis created real-time algorithm to integrate measurements from visual cues with measurements from onboard sensors to estimate the attitude position and velocity of a quad-rotor helicopter in a local navigation frame, a system model for the ARDrone, and a feed-back controller for the vehicle's heading. The ARDrone, by Parrot SA, is a low cost quad-rotor helicopter that comes equipped with a variety of sensors including a forward-looking high-definition camera. The vehicle is capable of using its onboard sensors to adequately constrain the errors for pitch and roll in all environments, however the yaw axis is still subject to drift. This work utilizes a RANSAC based vanishing point detection algorithm to provide a reliable heading reference and integrates the vanishing point based heading measurements with the system's onboard heading measurements through an extended Kalman filter. In addition to estimating the drone's heading, the Kalman filter also estimates the position and velocity of the drone as it moves through its environment. This system was able to provide a heading reference with an error of one degree for the drone and was shown to be capable of transitioning between vanishing points when the vehicle needed to change direction. The system also demonstrated that it was capable of generating an estimate of position and velocity. However because the position error was on the order of one meter, the estimate was not accurate enough for autonomous navigation.
DTIC Accession Number
Dean, James W., "Real-time Heading Estimation using Perspective Features" (2013). Theses and Dissertations. 860.