In this paper, we propose to use Conformal Geometric Algebra (CGA) to feature extraction and reduce dimensions of the data. First, the action data is preprocessed to normalize the data. Next, use CGA to reduce dimensions of data and create feature vectors. Finally, use the LSTM for training and prediction. The experiment was conducted on the CMU dataset with 8 different actions and the results showed that the proposed method has higher results than the previous methods. |