임베디드 시스템을 위한 LSTM-RNN을 이용한 Skeleton 기반 동적 제스처 인식

Abstract

In our work, dynamic gestures in SHREC’17 public database are recognized by extracting simple features from the coordinates of the hand skeleton and using a LSTM-RNN. By using our method, the higher recognition performance is acquired than the existing methods even if a simple structure of LSTM-RNN is used. The trained LSTM-RNN structure can be implemented to a embedded board because of its simplicity and small size.