We have proposed a 3D shape model retrieval system using physical blocks as a query. In this paper, we use not only the shape constructed by a user but also the its motion input by users as queries. The motions are measured with a gyroscopic sensor equipped in the base block and translated into several features such as average, variance, energy, frequency domain entropy, etc. We conducted experiments to evaluate the accuracy of model recognition and motion similarity among users and models using a classifier (Support Vector Machine). We also verified the effectiveness of adding motion into the shape retrieval through the analysis of retrieval results.
