Human Motion State Recognition Based on Multi-input ConvLSTM

Abstract

The human body generates acceleration signals during movement. After collecting and processing this signal, the movement state of the human body can be analyzed and the behavior of the human body can be judged. Human motion state recognition has a wide range of applications in fields such as health monitoring, somatosensory games, and user social behavior analysis. In this paper, the UCI data set collected by mobile phone sensors is used to build the model using convolutional long and short-term memory neural network (ConvLSTM) combined with multi-input CNN (Multi-head CNN), and combined with long and short-term memory neural network (LSTM) and convolutional memory Neural network (ConvLSTM) for comparative evaluation. The accuracy of the model in this paper reached 93.75%. Experimental results show that the new algorithm can more accurately classify and recognize the human motion state.