Abstract
Unlike the research of 2D tracks recognition, both techniques and concrete application examples of 3D tracks recognition are extremely few. Part of the reason of this is that the classification of 3D orbits is more complicated than that of 2D tracks. To overcome this problem, we previously examined the recognition of the sign language operation by Procrustes analysis. Procrustes analysis is a technique to compare two ordered data under various linear transformations which give flexibility to pattern deformation. However, it has some drawbacks in applying to pattern recognition problems; the transformation is asymmetry, and the definition of the prototypes is not clear. This paper explains our proposed methods to produce prototypes for Procrustes analysis, and the experiment which support them. To overcome these drawbacks, we propose some methods to produce prototypes for Procrustes analysis. We evaluated our proposed method by applying them to the spatio-temporal data of 11 sign language words, and we have achieved up to 87% recognition rate.
