Abstract
Image processing algorithms were developed and tested for computing a guidance signal for an agricultural tractor traveling on row crops. Field images of various row crops were recorded on videotape with a tractor-mounted camera. Taped segments were recorded on sorghum, cotton, and soybeans. Image quality was enhanced using an automatic-iris lens to control environmental light intensity variations. Also, an 850 nm, narrowband, bandpass optical filter was used to enhance reflectance difference between vegetative material and soil background. All computational image processing was performed on the videotaped data in a laboratory. Gray level images were partitioned into crop canopy and soil background components using the Bayes classifier. Image distributional properties were estimated by subsampling the gray level image. The centers of row crop canopy blobs on successive scan lines of the image plane were determined using the average of object-edge transition points computed from run-length encoding of the binary image. A heuristic classification algorithm was used to compute the lines passing through the canopy centers corresponding to crop rows. The algorithm was based on distance metrics relating crop canopy centers on successive image scan lines. A comparative measure of the heuristic algorithm performance in line detection was provided using the Hough transform. The primary limitation of the Hough method was the time required to compute the lines for real-time guidance. Image plane line parameters for crop canopy projections were used to compute the guidance signal, in terms of a tractor heading error and offset error, from geometric relationships between the plane of the field and the rigidly mounted camera. Results are presented on the individual processes of image subsampling, image thresholding, computing crop row centers, determining the equation of the lines passing through crop canopy blobs on the image plane, and computing the tractor guidance signal.
Reid, John Franklin (1987). The development of computer vision algorithms for agricultural vehicle guidance. Texas A&M University. Texas A&M University. Libraries. Available electronically from
https : / /hdl .handle .net /1969 .1 /DISSERTATIONS -747082.