Methods for Cropline Following

Abstract

The Demeter Project of the National Robotics Enpineering Consortium of Carnegie Mellon University seeks to develop a robotic harvester capable of guiding itself through fields of crop. A variety of techniques for guidance have been experimented with, from GPS to stereo vision. This report details efforts at guidance hy means of processing 2-D images taken from color cameras positioned on each side of a harvester. These efforts are meant to provide a low-cost and computationally inexpensive solution to the positioning problem. Finding the dividing line between cut and uncut crop in 2-D images, the cropline, is naturally formulated as an image segmentation problem. Although extensive work has been done in this area, most segmentation algorithms do not operate in real-time. Real-time in this context means cycling at 4-5 Hz, which is the approximate minimum cycling time to smoothly guide a harvester travelling at 4 m.p.h. Even with the rapid increases in the speed of computing hardware, sophisticated segmentation routines often take several seconds, if not minutes, to complcte. Therefore, a suitable trade-off between the reliability of results and the speed of the algorithm must be found. This report covers four methods for finding croplines. These are: a modified version of the algorithm presented by Ollis and Stentz [13]: a model-based color segmenter: a texture-based segmenter; and a color edge-detector. The speed and accuracy of these four algorithms are compared on images of alfalfa and Sudan in flat and bedded fields.