Electric Appliance Parts Classification Using a Measure Combining the Whole Shape and Local Shape Distribution Similarities

Classification of electric appliance parts is one of the interesting and practically valuable applications for 3D object recognition. Based on existing works, in this paper we try classifying electric appliance parts data obtained in an automatable process, which becomes a basis for automated recycling system. The dataset includes deformable objects such as cables as well as various rigid objects, some of which lacking a large part of the surface because of self-occlusions and materials of the parts. To realize high accuracy in classification, after the comparison of several similarity measures, we combine a measure which describes well the whole shape similarity with a measure that expresses the ratio of local surface patterns that appears in each model. The latter measure is suitable to describe the similarity of deformable objects that the whole shapes are heavily dependent on their configurations. We also investigate how the scale of computing local feature affects the classification result.