Abstract
As part of our research project on visual recognition of an autonomous agricultural tractor, we developed an image processing program that can detect on a real-time basis the movements of both tillage and no-tillage portions of an experimental field plowed by the tractor. Recognition is based on animation images taken by a video camera. Our analysis of static images, done for the purpose of verifying the reasonableness of parameters set to analyze animation images, allowed boundary regions to be recognized very clearly. The processing took about 0.1 second per static image. Analysis of animation images brought satisfactory field recognition results in both fine and cloudy conditions. Thanks to the use of algorithm incorporating the concept of region growing, our program was able to process animation images as fast as about 0.2 second per frame even for images taken in cloudy conditions, which have been difficult to recognize with traditional techniques.
