| With the rapid development of microelectronic technologies,the accuracy of sensors have been improved to a higher degree,which also makes the technology of human behavior recognition based on sensor signal is developing rapidly.Human behavior contains a wealth of information,which can fully reflect the movement and physiological state of the human body,so it is of great significance to the study of human behavior.Based on embedded technology and pattern recognition algorithm,a wearable recognition system is designed.The recognition system can collect real-time information,integrate various sensor information recognition,and finally realize real-time intelligent terminal display.Through the fusion and identification of human body surface electromyography(sEMG)and human body posture information,four human behavior states of walking,jumping,going upstairs and downstairs can be effectively identified.sEMG signals of human body are signals generated by contraction of human muscles.Muscle state is the best manifestation of current human behavior state and is the direct response of human body movement behavior.Human body posture information is indirect signals generated when human body moves,and contains space-time information.Therefore,We fuse and analyze the two types of information to identify human behavior.The main work of the paper is as follows:(1)Overall hardware system construction.After comparing the current mainstream research technologies,the use of wearable embedded devices can realize human freedom to the greatest extent.RaspberryPi is the control host,WIFI is the auxiliary communication method,through the computer interface technology,external sEMG sensor and human posture information sensor are connected to achieve the collection of various types of human data.(2)Multi-sensor data acquisition system.In this paper,two kinds of sensors are used to obtain sEMG signals and human body posture information.Multi-sensor data acquisition system is mainly composed of human surface electromyography signal acquisition module,human posture information acquisition module and data synchronization module.Aiming at the multi-sensor acquisition task,a concurrent data acquisition system was designed to realize the concurrent acquisition and synchronization process of the two sensor data.(3)Data preprocessing of the original signal.In this paper,SNR calculation is carried out on the collected sensor data to obtain the channel with higher ranking as the final identification object.At the same time,frequency domain analysis is carried out on the signal conditioned by hardware.It is found that some noise still exists in the signal,and digital filter is adopted to filter it.In order to realize the distinction between the human body’s active state and static state,that is,the distinction between the active segment and the static segment,the fuzzy entropy algorithm is used to realize the detection of the start and end points of the active segment.(4)Classification and recognition based on multi-sensor data fusion.In this paper,two kinds of sensors are used to extract features and fuse the features.PCA is used to reduce the dimension because the synthesized features have high dimension.At the same time,in order to realize the optimal classification,this paper compares the two classification schemes of BP neural network and support vector machine,and finds that the recognition effect of support vector machine for this system is better than that of BP neural network.(5)User terminal design.Aiming at the problem that system data cannot be displayed in real time,a user terminal is designed based on Qt.The terminal can realize real-time data visualization,system file transmission and other functions,and improve user experience. |
