Acceleration Gesture Recognition Based On Long-short Term Memory Network

Human-computer interaction is the way and means of information exchange between human and computer.It is a comprehensive interdisciplinary subject of cognitive psychology,ergonomics,multimedia technology and virtual reality technology.With the wide application of computers and various embedded devices in daily life,the traditional human-computer interaction method can not meet the requirements of people's intuitive and efficient information exchange,people began to seek a more natural and convenient way of interaction.Gestures can express people's needs intuitively and quickly.Gesture recognition with gestures as an interactive way begins to become a research hotspot of human-computer interaction.Gesture recognition is generally divided into vision-based gesture recognition and wearable sensor-based gesture recognition.The former is susceptible to illumination and background,and requires high quality image quality,which is not conducive to the popularity of gesture recognition.The sensor is small in size and high in sensitivity.It is less affected by the external environment during the acquisition process and is easy to be implanted into the wearable device.The gesture recognition based on the acceleration sensor becomes a new human-computer interaction mode.In this paper,the acceleration sensor is used to collect data,and the acceleration gesture recognition algorithm based on convolutional neural network and long-short memory network is studied.The designed system is verified by experiments and the effectiveness of the system is proved.The main research contents of this paper are as follows:(1)Sensor data acquisition and preprocessing.Wiimote is used to collect different gesture acceleration data and transmit it to the computer for storage in real time.Then,the collected gesture signal is smoothed and processed,and the improved swab algorithm is used to obtain effective gesture data,and a gesture data set based on the acceleration sensor is constructed.(2)Gesture feature extraction.The convolutional neural network is used to extract the gesture features,the training set data is used to train the network model,and the trained model is used to obtain the test set gesture features,and the feature information capable of effectively representing the gesture action is obtained.(3)Acceleration gesture recognition.According to the timing characteristics of gestures,an acceleration gesture recognition model based on long and short time memory networks is constructed.Acceleration gesture recognition is performed in conjunction with deep convolution features.This article uses the acceleration gesture dataset collected by Wiimote and the Actitracker public dataset for testing.The experimental results verify that the long-shortterm memory network method can effectively recognize gestures and improve the accuracy.The recognition rate of 96.5%and 98%is achieved in the gesture dataset and Actitracker dataset respectively.