LSD and Skeleton Extraction Combined with Farmland Ridge Detection

Abstract

In the traditional agricultural machine-vision navigation, Hough transform was widely used to carry on the straight-line detection of the ridge line. As for the high complexity of Hough transform, amounts of pretreatment on the images need be done to reach the effective results, however, the detection results are unsatisfied. Plentiful false positive and false negative exist during the detection, so, it is difficult to attain the working accuracy of unmanned agricultural machine and hard to be used in the actual conditions. LSD (a Line segment detector) algorithm is a partial optimal detection algorithm, which bases the pixels gradient and obtains straight line through the regional growth, it owns some advantages such as rapid calculating, better testing result, rare false positive and false negative, etc. This paper attempted to apply the LSD algorithm to the agricultural machine-vision navigation, and advance the detection accuracy and results by extracting the skeleton of crops images. By experimenting on four different crops, the result demonstrates that the LSD algorithm is better than the Hough transform on the accuracy and effect, meanwhile, it has the better engineering-applied value.