3D impurity inspection of cylindrical transparent containers

This paper presents a method for automatically detecting and three-dimensionally positioning particles based on sequences of 2D images of rotating cylindrical transparent containers. The method can be used in the manufacturing industry by distinguishing between particles residing inside or outside a closed container. A foreign particle on the inside is a cause of rejection, whereas a particle on the outside might represent harmless dirt. The proposed method detects, tracks and positions particles and classifies containers accordingly as either rejected or accepted. Experimental results have shown an accuracy of 85% on a test set of 20 containers and an average execution time per container of 1 s.