Research On Gesture Recognition Based On Wearable Accelerometer And Photoplethysmography Sensor

As a novel interface between human and intelligent system,gesture interaction which involves sensor technology and pattern recognition technology has been widely used in social life.Among them,wearable technology has become more suitable for gesture interaction system due to its convenience and practicality.In commercial wearable devices,accelerometer(ACC)and photoplethysmography(PPG)sensors are two mainstream sensors for gesture information capture.Research on gesture recognition is usually carried out in stationary conditions while users are often in different life scenarios,such as those for work or fitness.As a result,the accuracy of the gesture recognition will be affected by the motion noise.In order to explore the feasibility of using wearable devices to realize practical gesture interaction applications in different motion scenarios,this study conducts investigation on gesture recognition technologies based on ACC and PPG sensors.The main innovations and contributions are as follows:(1)According to the meaning of gesture,difficulty of execution and joints of hand,fourteen gestures are designed to constitute target sets.Fourteen participants are recruited and experiments are conducted in four different motion scenarios including sitting,walking,jogging and running.Four data sets involving ACC signal,red PPG signal,green PPG signal and infrared PPG signal are collected and constructed.(2)A gesture repetition segmentation method based on motion noise and PPG signal is proposed.Its effectiveness is verified in four motion scenarios,and the factors affecting the performance of the algorithm are explored.(3)The influence of motion noise on the quality of ACC and PPG signal is explored.By analyzing the signal-to-noise ratio(SNR)of hand gesture in different motion scenarios,ACC signal is found to be more sensitive to motion noise,while PPG signal is more insensitive.(4)By using the support vector machine(SVM),one-dimensional convolutional neural network(CNN)and long short-term memory(LSTM),the experiments on 14 gestures under different motion scenarios are conducted,and the performance of ACC and PPG signal on hand gesture recognition is compared from different perspectives.The experimental results demonstrate that,compared to ACC system,the PPG system obtains higher recognition accuracy and has the ability of reducing user's training burden,thus it is more suitable to realize the application of hand gesture interaction.In addition,the combination of ACC and PPG,as well as increasing the channels of PPG sensors are proved effective to improve recognition accuracy.