Research On Gait Recognition Of Indoor Positioning For People Based On MEMS Sensor

In recent years,Gait detection has been widely used in the fields of medicine,criminal investigation,navigation and position.In indoor positioning,real-time detection of pedestrian gait can provide a priori information to assist indoor positioning,which can improve the positioning accuracy of Pedestrian Dead Reckoning(PDR).In the narrow sense,the purpose of gait detection is to detect the static state of landing during walking.However,the gait detection is to recognize various gaits of pedestrians in this paper.With the development of Micro-Electro-Mechanical System(MEMS),the acceleration and angular velocity of human can be obtained for analysis and recognition.This paper recognizes the five gaits of walking,stationary,running,and up and down stairs,which is based on the MEMS inertial sensor.In response to this problem,the main research contents of this paper are listed as follows:1.This paper constructs the original input features for classification by analyzing different pedestrian gaits,and designs the preprocessing scheme.Firstly,Inertial Measurement Unit(IMU)is worn on the ankle of testers,and three-axis acceleration and three-axis angular velocity of stationary,walking,running,up and down stairs are collected by IMU.Then,it is validated by experiments that the acceleration signals and angular velocity signals obtained from these five gait patterns have certain differences,which can reflect the characteristics of each gait.Thus,three-axis acceleration and three-axis angular velocity data obtained from the sensors are taken as the original classification input features.Finally,some methods are used to preprocess original input features,such as smoothing filtering and data calibration.2.This paper proposes a gait recognition(GR)algorithm termed as FPRF-GR.Firstly,FPRF-GR uses windowing,coordinate system transformation,Fast Fourier Transform(FFT),and Principal Component Analysis(PCA)to construct features with the preprocessed data.FFT and PCA can eliminate redundant and defective features,which can improve recognition accuracy.Thus,this paper innovatively proposes a fusion feature engineering,which combines FFT and PCA.Moreover,in order to design an excellent classifier,this paper compares Random Forest(RF)with several commonly used classification algorithms on the requirements of accuracy,generalization ability,and speed and noise resistance of the gait model,and the comparison results show that the model constructed based on RF can meet the model design requirements for our experiments.Thus,FPRF-GR uses RF as the classifier,which is built according to the data generated by feature construction,and uses the 10-fold cross-validation method to evaluate the model.Finally,this paper proposes an optimization scheme for two parameters,including the number of decision tree and the number of samples in RF.3.This paper confirms the effectiveness of FPRF-GR,and combines it with hardware devices to test in real scenes.Firstly,the optimization scheme of the above parameters is verified.Then,the effectiveness of FFT+PCA fusion feature engineering in FPRF-GR is showed by the comparison with different feature engineering.Secondly,this paper compares FPRF-GR with several commonly used classification algorithms.The experimental results show that FPRF-GR can achieve an average classification accuracy to 98.2% for walking,running,stationary,up and down stairs,Receiver Operating Characteristic(ROC)curve is optimal and the model training time is the shortest.Finally,this paper combines FPRF-GR with hardware devices,and to be tested in real scenes with the accuracy of 98%.This paper realizes the recognition of pedestrian gait based on MEMS sensor,and the accuracy of recognition rate in real world scenarios is 98%,which indicates that the proposed algorithm has good practicability and broad application prospects in the future.