Computer vision assistance for guidance of agricultural tractors in varying crop environments

Abstract

Computer algorithms were developed for generating the guidance parameters necessary to steer an agricultural tractor in field operations. A variety of operations were considered in order that the guidance program be general enough for application to many typical field conditions including curved rows and a single edge. The main objectives of the study were to identify suitable image processing techniques to locate transitions in the field that could be used to generate guidance information, transform the transitions to relate to tractor position, determine appropriate techniques for computing the guidance information and to test the system on field images. A field scene simulator was developed to provide images of rowcrops as well as edges created from tillage operations. These images were used in the laboratory to gain an understanding of their influence upon image processing techniques as well as to provide a means for testing the ability of the algorithm to detect the transitions. Run length encoding was used to distinguish the crop canopy from soil background and to locate the edge in the image representing change in intensity level of disturbed soil, for the case of tillage applications. Three processing windows were placed over the desired edge or canopy and image processing was limited to these locations. Gram-Schmidt orthogonal methods were used to generate the least square polynomial passing through the resulting transition points. Points in the field were assumed fixed as the tractor moved over them. A guidance signal, composed of a direction vector and an offset error, was derived from the curve describing the data points.