Autonomous Ground Vehicle Navigation Using a Novel Visual Positioning System

In this paper a novel visual positioning system for slip correction of vehicles in GPS-denied environments based on Kalman filter is presented. Kalman filter fuses 2 sources of data; the first source of data is obtained from odometry, while a computer vision algorithm is used to obtain the second source of data. Using a plane-based pose estimation algorithm that employs four reference points and the Consensus-based Tracking and Matching algorithm (CMT) we were able to obtain pose estimation with centimeter accuracy. The four reference points correspond to the pixel coordinates of the corners of a rectangle closing a region in the roof where the camera should be pointing. Since the employed camera is mounted into the vehicle and it has eye fish lens, an image distortion correction method is used. As the CMT algorithm is robust to scale and rotations, position and orientation (POSE) estimation using computer vision is obtained even if the roof is not completely visible. Autonomous navigation of a ground wheeled vehicle was achieved using the proposed algorithm performing different courses on long period of time and no big slip was observed.