키넥트 센서 데이터를 이용한 K-pop 포인트 안무 분석에 관한 연구

Abstract

In this paper, we suggests a method of classifying Korean pop (K-pop) dances. In order to accomplish this, we construct a K-pop dance database, including 200 dance movements produced by four professional dancers and forty amateur dancers from skeletal joint data obtained by a Kinect sensor. First, we obtain thirteen core angles representing important motion features from 25 markers in each frame. The calculated angles are normalized with a min-max normalization technique and four statistical values (minimum value, average value, maximum value, variance) are calculated every 30 frames. The 13 statistical angles calculated through the above procedure are used as feature vectors by connecting them. Then, a dimensionality reduction is performed with a combination of principal component analysis and Fisher’s linear discriminant analysis. Finally, we design a Regularized-Extreme Learning Machine (R-ELM) classifier using ridge regression analysis. Experimental results show that classification performance is better than existing methods such as k-NN (K-Nearest Neighbor), SVM (Support Vector Machine) and ELM (Extreme Learning Machine).