Research On Indoor Location Algorithm Based On Machine Learning

As the rapid development of society,whether in the field of nation defense or people's daily life,the precise positioning of navigation technology becomes more and more important.In the outdoor environment,GPS(Global Positioning System)is the current technology which is a relatively mature and the most widely used in positioning and navigation system.In the indoor environment,especially in some public densely populated areas,the demand of location navigation,context awareness and real-time monitoring of people or things based on location information services,access to the indoor environment of people or objects is becoming stronger.High-precision indoor positioning technology has become an essential technology for people to get these location services.In this paper,we study the indoor location map matching algorithm,and re-improve the original map matching algorithm.The accuracy and robustness of the algorithm are verified by multi-person field experiments.This article first introduced a variety of indoor positioning technology,elaborated the positioning principle of each indoor positioning technology and compared its advantages and disadvantages.By using the factor graphs of the graph model,the sum-product algorithm and the max-sum algorithm,this paper builds the data fusion framework of intelligent autonomous learning function.Two sets of map matching algorithms were designed using Particle Filter(PF)and Conditional Random Field(CRF).This paper completed the data fusion of the inertial navigation trajectory and the map,improved pedestrian positioning accuracy.Based on the model analysis of actual pedestrian walking,the traditional particle filter map matching algorithm is improved.The map is discretized and the algorithm's memory overhead is reduced.A more reasonable weight transfer equation is proposed for indoor scenario,and more efficient particle reasonableness judgement algorithm is designed.In the CRF map match algorithm,we establish effective characteristic equations to improve accuracy and reduce energy consumption.Combining the eigenfunction of the fusion barometer data enables the model to use the elevator for floor switching.After completing the modeling of the linear conditional random field model.By establishing the training set data,the loss function was established by the maximum likelihood method,and the parameter training of the model was completed using the stochastic gradient descent method.The Viterbi algorithm was finally used to estimate the pedestrian's real-time location.Through the multi-person pedestrian test in the field,the positioning accuracy of the particle filter map matching algorithm and the conditional random field map matching algorithm designed in this paper has been greatly improved.The trajectory error of the conditional random field map matching algorithm is within five thousandths.The particle filter map matching algorithm has a trajectory error of less than seven thousandths.