Research On Indoor Positioning Method Based On Multi-type Fingerprints

With the rapid development of wireless communication technology and the Internet of Things,indoor positioning services have also been widely demanded.In addition,the rapid popularization of mobile intelligent terminals has made it very convenient to receive Wireless Fidelity,Bluetooth,and Magnetic Field fingerprints.Multi-type fingerprints positioning is studied in this thesis to realize the fusion positioning of Wireless Fidelity,Bluetooth and Magnetic Field fingerprints.Meanwhile,in view of the low positioning accuracy of the current location fingerprint positioning method and its inapplicability to multi-type fingerprints,two different positioning methods are proposed.The specific research contents are as follows.1.It combines the traditional machine learning algorithm with the deep learning model,and a multi-type fingerprints positioning method based on Multi-Task Learning and Weight Coefficients K-Nearest Neighbors is proposed.The Multi-Task Learning model fuses the features of different types of fingerprints to explore the potential relationship between them,so as to obtain the area label and physical coordinate of the target respectively.It also exploits the synergy between the tasks to improve the performance of the model.Then the label information obtained by Multi-Task Learning model is used to narrow the positioning range of the Weight Coefficients K-Nearest Neighbors,and another physical coordinate of the target is predicted in this range.Finally,the final position coordinate is obtained by fusing these two physical coordinates by the weighted average method whose weights are determined by the confidence provided by the positioning error prediction models.2.It converts multi-type fingerprints into graph structure data for processing,and a multitype fingerprints positioning method based on Graph Convolutional Network is proposed.The measurement point is the graph node,and the feature vector initialized by multi-type fingerprints is the node characteristic,so that the location network graph is established on the basis of an absolute rule and physical coordinates.Finally,the Graph Convolutional Network model is used to perform convolution operation on the location network graph to predict the area label of the unknown node.In addition,this thesis also explores the influence of the sparsity of the location network graph on the model performance and training time.These two methods in this thesis are tested on through the Miskolc Institute of Information Science Hybrid Indoor Positioning System data set,and compared with mainstream indoor positioning methods.The experimental results show that the two methods proposed can effectively solve the indoor multi-type fingerprints positioning problem.Compared with the first method,the second method has a stronger ability to predict area labels,but it also has certain application limitations.