Sistema de visão computacional aplicado à contagem de sementes de soja movimentadas por uma transportadora de sucção pneumática

Abstract

The seed counting systems available to the industry are still relatively expensive and have limited counting speed. One known difficulty is the challenge of segmenting objects when they are overlapping each other. This work proposes a counting system having a controlled environment in which soybean seeds are transported in an air conveyor pneumatic turbine and areinspected. The objects are segmented from the background,have thei rcontours detected and their convexhulls are calculated. There after the number of convexit ydefects is extracted, as well as other morphological characteristics of the object. Thus, a classic K-Meansclusterer is drawn, labelling each detected contour with the actual quantity of seeds. Finally the seeds are tracked by a frame-to-frame motion analysis. Although it produces accuracy that is still insufficient for industrial use, it constitutes a large throughput counting method with low overhead and is open for improvement in each one of its stages.