Body-relative navigation guidance using uncalibrated cameras
Author(s)Koch, Olivier (Olivier A.)
Massachusetts Institute of Technology. Dept. of Electrical Engineering and Computer Science.
MetadataShow full item record
The ability to navigate through the world is an essential capability to humans. In a variety of situations, people do not have the time, the opportunity or the capability to learn the layout of the environment before visiting an area. Examples include soldiers in the field entering an unknown building, firefighters responding to an emergency, or a visually impaired person walking through the city. In absence of external source of localization (such as GPS), the system must rely on internal sensing to provide navigation guidance to the user. In order to address real-world situations, the method must provide spatially extended, temporally consistent navigation guidance, through cluttered and dynamic environments. While recent research has largely focused on metric methods based on calibrated cameras, the work presented in this thesis demonstrates a novel approach to navigation using uncalibrated cameras. During the first visit of the environment, the method builds a topological representation of the user's exploration path, which we refer to as the place graph. The method then provides navigation guidance from any place to any other in the explored environment. On one hand, a localization algorithm determines the location of the user in the graph. On the other hand, a rotation guidance algorithm provides a directional cue towards the next graph node in the user's body frame. Our method makes little assumption about the environment except that it contains descriptive visual features. It requires no intrinsic or extrinsic camera calibration, and relies instead on a method that learns the correlation between user rotation and feature correspondence across cameras. We validate our approach using several ground truth datasets. In addition, we show that our approach is capable of guiding a robot equipped with a local obstacle avoidance capability through real, cluttered environments. Finally, we validate our system with nine untrained users through several kilometers of indoor environments.
Thesis (Ph. D.)--Massachusetts Institute of Technology, Dept. of Electrical Engineering and Computer Science, 2010.Cataloged from PDF version of thesis.Includes bibliographical references (p. 89-97) and index.
DepartmentMassachusetts Institute of Technology. Dept. of Electrical Engineering and Computer Science.
Massachusetts Institute of Technology
Electrical Engineering and Computer Science.