This project presents a robot navigation system
designated for testing a novel approach to self-location. The system consists
of a mobile robot with a parabolic camera mounted on and a PC Workstation
carrying out the computations. The parabolic camera provides the navigation
system with full 360° field of view images which serve as a feedback for a robot
trajectory correction. A robot self-location procedure uses a number of
specially designed locating objects called fiducials to determine the
robot position. The proposed fiducials are covered with a three
dimensional pattern which makes possible the pose-estimation without complex 3D
reconstruction methods as the currently available approaches require. The main
idea is that if an observer looks at such a fiducial from some unknown
angle he can easily figure out what this angle is. The only thing to do is to
calculate the average amount of light from the fiducial reaching the
observer. In terms of image processing this implies calculating an average
grayscale value over the fiducial in a picture taken by the observer.
Having at least two fiducials at his disposal the observer, therefore,
knows the angles at which he sees them and can estimate his position. In addition,
a single panoramic image obtained from the camera contains a complete scene of
the robot environment along with all of the fiducials. This feature of
panoramic images introduces further simplification into the self-location
algorithm. Some experiments were conducted and the corresponding estimation
results are presented. A shape of a mirror which maximizes the fiducials’
resolution on the image plane was calculated. Such a mirror is optimal for the
presented pose estimation algorithm and can be used to enlarge the robot working
area.
