Development of Motion Analysis and Classification Using ELM and Autoencoder
Donghyun Kim · 2017
This paper is concerned with a development of motion analysis and motion classification based on ELM (Extreme Learning Machine) and AE (Auto-Encoder) for a Korean pop (K-pop) dance classification. First, we calculate 13 angles representing important motion features from 3D marker information obtained by Kinect camera. The statistical values of these angles are concatenated with feature vectors for all of the frames of each point dance. Second, we perform a dimensionality reduction based on PCA (Principal Component Analysis) and LDA (Linear Discriminant Analysis). Finally, we design ELM or AE classifiers for dance motion classification. For this, we construct a K-pop dance DB with 800 and 400 dance-movement data points including 200 dance types produced by 4 professional dancers and 40 trainees for training and testing data set, respectively. The experimental results revealed that the presented ELM method showed a good classification performance in comparison with KNN (K-Nearest Neighbor). For further research, we shall perform the experiments based on AE.