Application Of The Hough Transform For Seed Row Localisation Using Machine Vision

This paper compares two methods based on machine vision to provide driver assistance in seed drill guidance in order to improve spacing accuracy during contiguous passages.
The first case consisted of following the furrow created at the preceding passage by a special marker disc attached to the seed drill. A camera was located on the tractor and detected this furrow. In the second case, the seed rows themselves were detected by the camera without making use of the marker disc.
In both cases, several video sequences were acquired in various situations, including different soil textures and various illumination conditions (375 sequences were acquired during three years). A pre-treatment of these sequences was performed and included a background subtraction in order to remove shadows and other wide unevenness. In the first case, the best results were obtained by using an image treatment based on the Hough transform coupled to a recursive filter. The search of the maximum of the Hough transform was performed using a mean shift algorithm. In the second case, where several parallel rows were simultaneously present on the images, an adapted Hough transform was proposed which took into account the a priori knowledge of the rows spacing. The trueness and precision in row detection were superior in the second case. The results are compatible with the application, since the trueness was smaller than 30 mm. This suggested that it can be possible to assist the manual guidance of a seed drill by an automatic system consisting in a camera detecting the seed rows.