Multiple RNN Method to Prediction Human Action with Sensor Data

Abstract

Human body motion includes the complex spatiotemporal information and human body motion prediction is useful in the human-computer interaction. An Encoder-Multiple-Recurrent-Decoder (EMRD) model to learn human action from sensor data and predict the later ones is proposed in this paper. The kernel of this method is recurrent neural networks (RNN). The model is used to predict the next several frames of a set of sensor data, which is continuous data but is pre-processed by embedding method proposed in this paper. EMRD extends the previous Encoder-Recurrent-Decoder (ERD) models and Long Short Terms Memory (LSTM) model which are used in the video human body movement prediction.