Vision-based crop line tracking for harvesters

Abstract

A self-propelled harvester is provided with apparatus for tracking a crop line and steering the harvester along the crop line without operator assistance. The apparatus includes a video camera for viewing the landscape in front of the harvester and forming an image which may be electronically scanned pixel-by-pixel and line by line to produce a video signal containing information regarding the spectral characteristics of the image at each pixel. A microprocessor is programmed to act as a discriminator and a segmentor. The discriminator produces discriminant signals D(i,j) where i is the magnitude of a discriminant function and j is the relative position of a pixel on a scan line. The segmentor determines a step function which best fits the magnitudes of the discriminant signals for one scan line. The pixel position j.sub.n corresponding to the step of this step function is taken as a vote on the location of the crop line and thus a vote on the direction the harvester should be steered. The votes are summed for each pixel position j over one full scan of image. The summed votes are then normalized, time-weighted, and averaged with the normalized and time-weighted votes derived from previously scanned images to derive a steering command.