The Research On Single Soldier Body Posture Recognition Technology Based On Body Area Network

The wireless body area network(WBAN)can monitor human vital signs,posture characteristics,surrounding environment and other information in real time,and its civil value is often reflected in health monitoring and other scenarios.In the military,the individual body network is used to collect body motion recognition by collecting information such as the acceleration of the warrior.It can be used for the evaluation of injuries in the limbs during wartime,as well as for the follow-up of daily tactical action training and auxiliary equipment such as exoskeletons.Therefore,the value of military research is also very prominent.Based on the military application background,this paper focuses on the related technologies of body posture classification and recognition based on MEMS inertial sensors and WBAN,and mainly carries out the following work:1.In this paper,the establishment and network access,data acquisition and network communication of WBAN based on the Zigbee communication protocol are studied.The wearable data acquisition system is designed by using the nine-axis acceleration module JY901 and four modules with Zigbee protocol stack.The acceleration module is clamped to the neckline,and the signals of the 10 postures commonly used by the soldiers are collected,including the three major steps of the queue(step walking,goose step,running),common tactical movements(low posture,high posture),physical training(push-ups,sit-ups),going upstairs and going downstairs.2.For the collected nine-axis body movement data,by comparing the characteristics of each dimension signal,considering the retention of the signal characteristics of each dimension while reducing the amount of calculation,a method that the acceleration vector and amplitude(or vector amplitude)and angular velocity vector and amplitude are processed in parallel is proposed.3.The preprocessing methods such as signal denoising and normalization are studied.The characteristics of the signal are analyzed from the three domains of time domain,frequency domain and wavelet,and the feature vectors of each domain are integrated to extract the high-dimensional mixed feature vector and feature vector space.4.For the problem of "dimensional disaster",this paper uses linear discriminant analysis(LDA)to reduce the dimension of the feature vector,maintains the body recognition rate and reduces the computational complexity of the algorithm,which improves the recognition efficiency of the posture greatly.5.Finally,this paper studies the machine learning algorithm,and proposes a support vector machine(SVM)algorithm based on hybrid features,combining mixed feature vector and LDA.We use the feature vector space to train vector machine and to realize the recognition and classification of ten body modes,and obtain the ideal recognition rate.