Research And Implementation Of Attitude Recognition Algorithm Based On Multi-sensor Data Fusion

With the continuous development of artificial intelligence and Internet of Things technology,research in various fields has made great progress.Attitude recognition technology is closely related to daily life,and it has attracted more and more attention.Many jobs also increasingly require convenient attitude recognition equipment in life.On the one hand,attitude recognition equipment can identify some dangerous actions to avoid accidents and ensure the safety of person;on the other hand,by collects various attitude data and analysis worker behaviors can effectively improve work efficiency.For this reason,this paper designs an attitude recognition device based on multi-sensor data fusion.Compared with traditional recognition devices,it has the advantages of small size and convenient carrying,and is more popular in daily life.First,the hardware of the attitude recognition device includes the main control module and the attitude module.Among them,the attitude module is used to collect human attitude data,and the main control module is used to receive data collected from the attitude module.The attitude module uses the nine-axis sensor ICM-20948 chip,which consists of a three-axis accelerometer,a three-axis gyroscope,and a three-axis magnetometer.The main control module uses MT2503 chip,which is used to receive the data of the attitude module and send the data to the server.Secondly,according to the shortcomings of inertial sensors,three sensors need to fusion and attitude calculation.After analyzing a variety of commonly used data fusion algorithms and attitude calculation algorithms,select the quaternion method and extended kalman filtering algorithm.The fusion algorithm uses the data solved by the gyroscope as the equation of state,and the data measured by the accelerometer and magnetometer as the measurement equations.It is used to correct the angle calculated by the gyroscope.Then,use the attitude module to collect different attitude of different people,such as running,walking,going up and down stairs.According to the data signal processing methods,such as FFT transform,wavelet transform.The data is compared and analyzed in the time and frequency domain,and extracted the characteristics of each attitude,which lay the foundation for subsequent attitude recognition.Finally,in order to accurately identify and classify actions,select an EM algorithm based on a Gaussian mixture model.The EM algorithm is an iterative optimization algorithm.Iteration is divided into an expected step and a maximum likelihood step.According to each iteration step,the optimal convergence value can be effectively found.The eigenvalues of each attitude action are sent to the algorithm for model learning.Then test the trained model,use the collected sets of attitude data and algorithm model to identify and classify it,observe its recognition accuracy,and continuously optimize the model to achieve accurate recognition of human poses.