Visual navigation for an autonomous mobile vehicle

Abstract

Image understanding for a mobile robotic vehicle is an important and complex task for ensuring safe navigation and extended autonomous operations. The goal of this work is to implement a working vision-based navigation control mechanism within a known environment onboard the autonomous mobile vehicle Yamabico-II. Although installing a working hardware system was not accomplished, the image processing, model description, pattern matching, and positional correction methods have all been implemented and tested on a graphics workstation. A novel approach for straight-edge feature extraction based upon a least squares fitting of edge-related pixels is presented and implemented for the image processing task. A simple method for determining the camera's location and orientation (pose) follows by matching the vertical extracted edges from an image with the linear features of a two-dimensional view of the modeled environment based upon an estimated pose of the robot. Image processing, construction of the two-dimensional view of the model, and pose determination are conducted sequentially in less than one minute for a 646 x 486 pixel image on a 35 MHz processor. The pose determination results have been tested to be accurate to within a few inches for translational error and within one degree rotational error.