Detection of defective products using stereovision

In this paper a stereovision detector of defective products for vision inspection on the production line is proposed. The proposed stereoscopic classifier uses an artificial neural network (ANN) to classify products by analysis of images taken from both the left and right view. The detector may be dedicated and tuned to a given product although it is, in general, universal as it can detect various defects in various products. Experiments conducted with plastic elements taken from the real injection molding production line confirmed that the proposed detection system operates correctly with both modes (dedicated and universal), achieving sensitivity about 90 % with the relatively simple ANN. In all tested cases, the stereovision-based solution offers higher sensitivity of defects detection than the classic monovision solution.