Affiliations: [a] School of Computer and Communication, Lanzhou University of Technology, Lanzhou, China
| [b] Information Center, Lanzhou University of Technology, Lanzhou, China
Abstract: Smartphone has been used for recognizing the different motion activities. However, current studies focus on either improving algorithm factor or adjusting neural network structure factor rather than on time cost factor and actual application factor. A novel method to consider these four factors comprehensively enhancing recognition of motion state accuracy is proposed. An architecture of the Bi-LSTM neural network and the TensorFlow machine learning system are used to classify the motion state and evaluate its experimental results. In addition, the Bi-LSTM neural network is compared with other neural network structures. Meanwhile, using the data captured by the accelerometer sensor and gyroscope sensor of the smartphone tests the Bi-LSTM neural network model. Experimental results show that using Bi-LSTM neural network and TensorFlow machine learning system to extract motion state characteristics, this method makes the motion state identification achieve 86.7% accuracy and the Bi-LSTM neural network model is better than other neural network models considering above four factors. The model of Bi-LSTM neural network can be used for other time-series fields such as signal recognition, action analysis, etc. This study provides a new method, which considers the four factors, to enhance the accuracy of the motion state classification.
Keywords: Deep learning, Bi-LSTM neural network, motion state, sensors of smartphone, TensorFlow
