Research On Indoor Positioning Algorithm Based On Machine Learning And Multi-source Information Fusion

With the rapid development of user-location-based smartphone services,indoor positioning technology has gradually become a research hotspot in the field of positioning technology.Due to the complexity of the indoor environment,the propagation of wireless signals indoors will be blocked by indoor obstacles.The accuracy of satellite positioning in the indoor environment is far less than that in the outdoor environment,so other sensors need to be used for indoor positioning.This thesis proposes a fusion positioning algorithm based on WiFi location fingerprint and inertial positioning,and uses machine learning algorithm to optimize it,and finally achieves good results.The research work of this thesis is mainly divided into the following three parts:In the first part,this thesis studies and proposes an indoor positioning algorithm based on deep learning and WiFi location fingerprinting,which maps WiFi access points to vectors to learn the location relationship between WiFi access points.Experiments on a large indoor positioning dataset show that this method improves the positioning accuracy by 8.1%compared with the traditional WiFi location fingerprint positioning method.In the second part,this thesis studies and proposes an inertial positioning algorithm based on deep learning.Compared with the PDR algorithm,the inertial positioning algorithm improves the prediction accuracy of user displacement by 26.7%,and does not need to set different parameters for different pedestrians like the PDR algorithm,and does not need to be set for different pedestrians like the PDR algorithm.different parameters.In the third part,based on the first two parts,this thesis designs an indoor fusion positioning algorithm based on WiFi location fingerprint positioning and inertial positioning.The algorithm uses the WiFi location fingerprint model to predict the absolute position of pedestrians indoors;the inertial positioning is used to predict the relative displacement of the user in a given time.Finally,a joint loss function is designed to make the final prediction result combine the above two parts of prediction information.The experimental results show that the fusion positioning combines the advantages of WiFi fingerprint positioning for accurate indoor position prediction and inertial positioning for accurate user relative displacement prediction.Compared with using single positioning technology,the prediction results of pedestrian paths are more accurate and robust.The positioning accuracy is improved by 41.5%compared to using only WiFi location fingerprinting.In the positioning stage,this thesis also proposes a correction method for prediction results based on building information.According to the design drawing of the building,the location where the user should not appear in advance is marked.If the predicted result of the pedestrian track appears in an unreasonable location in the building,the post-processing correction algorithm designed will correct the predicted coordinates to a reasonable location.