Indoor Positioning Algorithm Based On Deep Learning Of Walking Motion State

With the development of the Internet and smart terminals,location-based services are becoming more and more widely used in people's daily lives.Due to the comprehensive coverage of the global positioning system,the development of outdoor positioning technology is quite mature.For indoor development,as the building blocks the GPS signal,the outdoor positioning method is not applicable indoors,and the existing indoor positioning technology is not mature,indoor positioning technology is not widely used in the market at present.In this context,this paper selects the inertial sensor and PDR positioning technology as the research basis,which is in order to solve the problem of indoor positioning,and corrects the positioning error that existed in the inertial sensor positioning technology process,improves the indoor positioning accuracy as well.In order to correct the indoor positioning error,the inertial sensor data of indoor pedestrian movement is classified into motion state in this paper,and the predictive function of the conditional random field algorithm is used to combine the pedestrian motion state with the pedestrian dead reckoning technique.The main research contents of this paper include the following aspects:(1)Constructing a deep learning model to predict the state of pedestrian walking in indoors.In this paper,indoor pedestrian walking status is divided into five states: upstairs,downstairs,elevator,normal walking,turning,and the sensor data which is collected by a mobile App.The data set is divided into training set and test set after preprocessing,which is used for deep learning recognition model training and testing.The deep learning model was tested by a test data set with an accuracy of 0.97.Finally,the paper transforms the deep learning recognition model into a predictive model for the prediction of indoor pedestrian walking state sequence.(2)Constructing a conditional random field model to predict the pedestrian walking path.According to the prediction principle of Viterbi algorithm in CRF,this paper constructs a CRF walking state prediction model,and combines the map matching principle to construct indoor maps,set indoor map nodes,and match the walking motion state sequence with indoor mapnodes for predicting indoor pedestrian walking paths..The model construction process is implemented by Python code,and the model input is a walking action state sequence,and the output is a walking path represented by a map node.(3)Constructing a pedestrian dead reckoning algorithm structure to restore pedestrian indoor walking trajectory.Based on the theory of PDR positioning algorithm,this paper analyzes and processes the indoor walking sensor data by writing matlab program,and obtains pedestrian walking steps,step frequency and step length and heading angle.The walking trajectory is matched with the indoor electronic map through the zero-crossing data processing and coordinate conversion method,and the walking trajectory recovery situation is displayed on the indoor map.In order to solve the positioning error and navigation trajectory drift of PDR algorithm,this paper combines CRF and PDR algorithm to correct the navigation trajectory,reduce the cumulative error and improve the positioning accuracy.(4)Experimental verification of the indoor positioning method mentioned in this paper.The experiment consists of four indoor walking routes.The experimental site is the first floor and the 14 th floor of the Science and Technology Building of Shen Zhen University.Before the experiment,an indoor electronic map was constructed based on the actual situation of the experimental site.This paper compares the PDR positioning effect with the pedestrian PDR+CRF positioning effect by importing the experimental data and navigation path into the map.In this paper,the PDR and PDR+CRF walking trajectory coordinates are compared with the real trajectory coordinates.The PDR positioning accuracy is 4.43,and the PDR+CRF positioning accuracy is 1.19.