Orientation Invariant Surface Classification Using Uncertainty Level Estimation

Abstract

This paper proposes and describes orientation invariant surface classification system used for connectors labelling. The presented system was tested by classifying sides of 8 different electrical wire connectors for automated ink-jet print labelling. Connectors were randomly placed on the conveyor, and the system identified their visible sides regardless of the orientation and camera’s viewpoint variation. All connectors in one batch were of the same type. Identified wire connector information was further fed to the industrial robot arm, which then could take free oriented connector and face it to the ink-jet printer on the required side. For classification task, four different classifiers were experimentally compared: Artificial Neural Network, Decision Tree, K Nearest Neighbours, and Quadratic Discriminant Analysis. The first order statistic and Scale-invariant feature transform were used for feature extraction from the images. The approach proposed allowed identification with 100º% accuracy depending on the selected level of uncertainty. Experimental results have shown that quantity of unrecognized samples for most connectors varied only in the range of few percents.

DOI: http://dx.doi.org/10.5755/j01.eee.19.10.5885