Research On Indoor Inertial Navigation Positioning Based On Deep Learning

As the time required for people to conduct indoor activities in modern society continues to increase,people's demand for location services is also increasing.Due to the obstacles of the building and the complex and changeable indoor environment,the satellite signal cannot support the indoor positioning requirements.Therefore,the Pedestrian Dead-reckoning(PDR)positioning technology that does not rely on external signals has received extensive attention and research.PDR positioning technology no longer relies on external equipment to provide support,and only relies on the data collected by its own inertial sensors,so the service quality of indoor location services is greatly improved The main content of this paper is to improve the step length estimation and heading angle estimation in PDR positioning technology.This paper first introduces the deep learning method to the step length estimation,and proposes two step size estimation algorithms based on recurrent neural networks,one is an RNN model based on single-step historical data,and the other is an RNN model based on multi-step historical data.Experiments show that the average errors of these two models are smaller than traditional estimation algorithms.It fully shows that the deep learning model has certain advantages in step size estimation.This paper proposes an angle correction algorithm,which is an improvement of the traditional heading angle estimation algorithm.In the angle correction algorithm,we first train a heading angle change value classifier through the deep learning method,and then use the classification result of the classifier as the innovation in the Extended Kalman Filtering process to iterate to obtain the best estimate of the current heading angle.Experiments show that the angle correction algorithm can effectively solve the problem of small heading angle estimates,thereby reducing estimation errors.The algorithm does not rely on external observations,but uses a deep neural network to fit its own data to provide additional information for the system.This method solves the problem of non-foot pedestrian dead reckoning system can not effectively seek observables.Finally,this article combines the improved step length estimation and heading angle estimation to conduct a comprehensive experiment of indoor inertial positioning.This experiment sets up three sets of control experiments,and analyzes the RNN-based step length estimation algorithm and angle correction algorithm from multiple aspects.The experiment shows that the improved Pedestrian Dead-reckoning method can improve the accuracy of pedestrian position estimation.In this paper,the deep learning method and the Pedestrian Dead-reckoning positioning technology are combined to effectively reduce the positioning error of the Pedestrian Dead-reckoning.