| An algorithm to extract the guidance parameters, vehicle offset and heading angle, from crop-rows digital images was developed and tested. The algorithm approached the guidance parameter determination as a pose recognition problem. Crop rows were treated as an object for detection that changed in pose depending on the vehicle guidance parameters. Principal component analysis was used to encode the original image space into a lower dimensional space (eigenspace) and then the pose-guidance algorithm (PGA) accomplished a classification of the encoded image. Two architectures were proposed for PGA, global PGA and angle PGA, and two kinds of classifiers were evaluated in this research: minimum Euclidean distance classifier (MEDC) and artificial neural network classifier (ANNC).; Three different camera heights and camera tilt angles were used to collect simplified images that representing the crop-rows scene for different vehicle guidance parameters. Using the MEDC, the best PGA performance was with the camera height and tilt angle of 1.22 m and 15 degree, respectively, using the half bottom of the image as ROI for global PGA and combination of this ROI and half top of it for angle PGA. At these settings, the average absolute errors between the actual and global PGA estimated pose were 0.5 degrees and 4.9 cm for heading angle and offset, respectively. For angle PGA, the average absolute errors were 0.5 degrees and 4.1 cm.; An ANNC was shown to be an option for the classification task of PGA.; Images representing five different stages of development for corn and soybean were collected. The PGA performance in both kind of images, raw and binary, was similar. Absolute error between the actual and the estimated heading angle ranged from 0.3 to 0.7 degree and from 0.3 to 0.9 degree for corn and soybean plants, respectively, depending on the stage of development. Absolute error between the actual and the estimated offset ranged from 2.2 to 3.3 cm and from 2.9 to 5.0 cm for corn and soybean plants, respectively, depending on the stage of development.; PGA performance for estimating heading angle was considered satisfactory. However, for offset estimation, PGA still requires some improvement. |
