情報処理学会 インタラクション2015

In this paper, we propose a rapid motion retrieval technique using Dynamic Time Warping.

Frames of motions are represented by feature vectors whose elements are integer values. The number of the

feature vector dimension is reduced by using the Principal Component Analysis method and values of ele-

ments in the vector are quantized into two bits. The similarity matrix between frames of motions represented

by the feature vectors is generated for rapid calculation of Dynamic Time Warping. Preliminary experiments

are conducted to nd the optimum dimension number of the feature vector by evaluating the motion retrieval

performance. Comparative experiments with existing methods have proved that our proposed technique can

complete retrieval tasks more than 30 times faster than the traditional Dynamic Time Warping method and

shown almost the same level of precision and rapid calculation time as the method described in [1] using the

k-d tree algorithm.