Gait Detection In Multi-state Based On Smartphone IMU

With the construction of smart cities and the development of Internet technologies,people have higher and higher requirements for positioning accuracy and practicability in location services.The improvement of satellite navigation system(GNSS)has been able to meet users' the positioning requirements in outdoor environment.How to provide users with location information in complex indoor environment has become a hot spot in the positioning navigation industry.Pedestrian dead reckoning(PDR)is an essential technology for positioning and navigation in today's complex indoor environments.As an important step in the PDR process based on motion sensor,the accuracy of gait detection and the reliability of step estimation directly affect the positioning accuracy of the whole system.Based on this,this paper took the ordinary smartphone inertial measurement unit(IMU)as the research object and did the following research:1.The performance of smartphone inertial sensor was analyzed.Aiming at the problem of low precision and high noise,a fusion noise reduction algorithm based on weighted moving average filtering and Kalman filtering was proposed to improve the reliability of step detection.2.This paper analyzed the characteristics of the acceleration waveforms of different pedestrian movement modes and different carrying positions of mobile phones,determining the step detection algorithm under different states,obtaining the corresponding optimal parameters of step detection,and improving the accuracy of step detection.A step size estimation model based on real-time stride frequency was established by detecting the real-time stride frequency of pedestrian.3.After comparing and analyzing the characteristics of inertial signals in different motion modes of pedestrians,this paper proposed the discrimination method with pedestrian's static and motion state,which based on acceleration threshold and window standard deviation,and the detection deviation of static and motion was controlled within 2%.By analyzing the real-time step frequency of pedestrians,the recognition method of pedestrian motion mode based on real-time stride frequency was proposed,and the discriminating error of the three modes of slow walking,normal walking and fast walking was less than 5%.4.After studied on the inertial signals of pedestrians traveling in different carrying positions,this paper formed the feature matrix by using the statistical algorithm to extract signals' time domain features.Then,this paper used the principal component analysis algorithm to select the features that best reflect the carrying position of the mobile phone.Afterwrds,this paper used the random forest algorithm to train and model the selected feature,and the model had a discrimination accuracy of 99%on the carrying position of the mobile phone.After adding the position recognition algorithm,the step detection accuracy was improved by 6.89%,and the distance estimation accuracy was improved by 7.65%.5.This paper extracted the time-frequency characteristics of the inertial signal when the pedestrian was in the backward state by using the empirical mode decomposition method.The maximmum correlation minimum redundancy algorithm was used to select the feature data that best reflected the pedestrian motion state.Moreover,this paper used the support vector machine to train and model the selected feature.The model had a precision of 96%for the pedestrian forward and backward state.After adding the recognition algorithm,the estimation accuracy of pedestrian travel distance was improved by 26.06%.