Building perspective models to guide a row crop navigation vehicle

To fully automate some in-field agricultural tasks, an autonomous vehicle has to be guided along a crop held. In this work, we describe how a vision system extracts the guidance information for vehicle navigation in a crop row. The method is based on extraction of the scene structure from perspective information that a set of coplanar parallel lines, representing the crop rows, generate in the image. To solve some drawbacks caused by these natural environments, a method to extract the row lines is described, and an approach to calculate the vanishing point proposed. A technique to recover missing information is also introduced. The method has been applied to long sequences of images taken by a moving vehicle in crop row scenes.