Research On Pedestrian Autonomous Positioning Technology Based On Inertial Navigation

With the development of technology,people require more detailed and specific service of positioning and navigation.Indoor positioning technology can meet people's demand for indoor navigation and fire rescue.As a low-cost indoor positioning technology without external signals,inertial navigation technology has attracted people's attention.However,the traditional foot-mounted inertial navigation technology has the problems of rapid error accumulation and trajectory divergence,which cannot meet the positioning requirements.In order to effectively improve the reliability and effectiveness of pedestrian autonomous inertial navigation technology,this paper designs a new pedestrian autonomous positioning algorithm based on the in-depth analysis of the error dispersion of the inertial navigation system,and introduces weighted time constraints into the footstep detection link,introduces a heading matching algorithm into dead reckoning,and uses convolutional neural network to complete the step size estimation.The main research content of the paper includes the following parts:1.In the process of using vertical acceleration to detect footsteps,the paper introduces weighted dynamic time constraints,supplemented by acceleration difference constraints and peak symmetry constraints,which effectively improves accuracy of peak detection.The paper also uses a low-pass filter to further reduce high-frequency noise interference in acceleration.2.The parameters of traditional step size estimation algorithms are difficult to determine,the functional relationship between parameters and step size is difficult to find,and the values of hyperparameters in the formula are difficult to determine.In this paper,convolutional neural networks are used in the step size estimation.By training the neural network to learn the step size features in the sensor data,the reliability and robustness of the step size estimation are improved.Because there is no suitable open source data,the paper also uses the inertial navigation module to collect,process and produce the test and training data sets required for the experiment.3.Aiming at the problem of insufficient utilization of heading information in the traditional STH(straight line trajectory update)algorithm,a heading matching algorithm is introduced.By using a heading with higher confidence to correct the heading at the current moment,the reliability of dead reckoning is further improved.Aiming at the misdetection of traditional zero-speed detection algorithms using angular velocity or acceleration modulus detection,the paper also uses generalized likelihood detection to enhance the difference between the zero-speed interval and the non-zero-speed interval,and improve the accuracy of zero-speed detection.The results of experiments using collected data show that the positioning error of the paper method is about 2.1% in long distance walking.