Research On Human Gesture Recognition Based On Wearable Devices

With the development of society and science and technology,intelligent products such as smart phones and watches have become essential daily necessities for people.At the same time,people also put forward higher requirements for their own and their family's physical health and exercise status.At present,the study of human posture recognition based on smart wearable devices is still in its infancy,and the research results are relatively few.Based on this,this dissertation takes the recognition of human motion gestures as the research object,and focuses on the collection of acceleration time series data,data preprocessing,feature extraction,and the recognition method of human motion gestures based on acceleration signals through intelligent wearable devices during human motion.Focused on the study.The main research content and author work of this article are as follows:1.In order to satisfy the research requirements of human motion gesture recognition based on time series,this paper introduces the data acquisition scheme and feature extraction using acceleration sensor and summarizes the commonly used human motion gesture recognition method through a large number of literatures.2.Propose human posture recognition using the similarity measure of acceleration time series data based on DTW distance,that is,calculate the similarity between time series of acceleration data of each human motion gesture,and finally combine the most common motions of the nearest neighbor algorithm to the human body.Gestures are identified.Experimental results show that this method achieves a good recognition effect.3.In order to solve the problem that the distance metric calculation error between time series is too large and the invariability of the time-series zoom translation is guaranteed,a shapelet feature extraction algorithm based on DTW is proposed,which combines the distance between DTW measurement time series and shapelet feature extraction.And experimentally validated seven different data sets,all with good accuracy.4.A LSTM-based human motion gesture recognition model was designed.The common motion sets in the six kinds of human daily life contained in UCI human motion data were used to transform the data set using shapelets feature extraction method.Finally,the model was modeled using TensorFlow platform.Build and train.The experimental results show that the LSTM-based human motion gesture recognition model has achieved good recognition effect for 6 types of movements inhuman daily life,and has achieved 100% recognition accuracy for the special fall state on the data set used.It is illustrated that the human posture recognition based on LSTM in this chapter has significant effects and certain practical significance.