Research On Human Posture Capture And Behavior Recognition Based On Inertial Sensor

With the development of science and technology,human body posture capture technology is widely used in various industries and has huge research value and development potential.Inertial human body posture capture technology has huge advantages over other implementation methods,so it has become a research hotspot in the field of human body posture capture.Therefore,this thesis designs a set of human posture capture system based on inertial sensors,and conducts research on human behavior recognition on the basis of this system.This thesis first analyzes the human body posture capture system and determines the various components of the human body posture capture system.In this thesis,15 posture measurement nodes are used to collect the motion parameters of each torso of the human body,and then perform posture calculation to obtain the corresponding posture data.Through wireless communication,the posture measurement node transmits to the wireless receiving node,and then the wireless receiving node passes the serial port communication method and transmits the data to the PC to display the posture of the human body.This thesis compares the selection and comparison of the inertial sensor module,wireless communication module and battery module of the attitude measurement node,then chooses to use the attitude calculation algorithm based on the gradient descent method to calculate the attitude,and calibrates and compensates the three kinds of sensors in the inertial sensor to improve the accuracy of attitude calculation.Then,according to the human kinematics and human level analysis,the 3D Max software is used to establish the human body model,and then the human body posture display is developed based on the Unity software.After testing,the human body posture capture system based on inertial sensors can better restore the human body posture.Based on this inertial sensor-based human posture capture system,this thesis designs 8actions to collect data samples.This thesis first uses the quartile method to eliminate outliers,and then performs temporal feature extraction.Aiming at the problem of too large feature space dimensions,principal component analysis is used for feature dimensionality reduction,and then SVM classification algorithm is used for human behavior recognition.After the construction,the average recognition rate of the 8 actions by the SVM algorithm was 92.5%.