CSI Open-set Identification And Positioning Method For Indoor Terminal Positioning

Indoor terminal positioning related technologies will be used in application scenarios such as the operation and maintenance of indoor robots in modern smart cities and the indoor path navigation requirements of personnel.Currently,using CSI fingerprints as Wi Fi signal samples for indoor positioning is one of the important methods in the field of indoor positioning.In order to ensure the positioning accuracy of this method,the workload of pre-acquisition and deployment is often very large.On the one hand,in order to simplify the deployment work,reducing the number of collection points will lead to missing fingerprints and form an open space.On the other hand,CSI is seriously disturbed by the complex indoor environment,and the samples to be tested at the collected points will also dynamically change to an open set.In summary,the above two CSI open set seriously interfere with the CSI fingerprint method,and lead to difficulties in long-term maintenance of the model.Therefore,there is a lack of a method for identifying and locating CSI open set.In this regard,this thesis proposes a CSI open set identification and positioning method for indoor terminal positioning under the non-global CSI acquisition method,which aims to reduce the workload of fingerprint deployment and reduce the risk of open space caused by CSI open set.First of all,to solve the problem of heavy deployment workload of the fingerprint database,this thesis designs a regression estimation method of indoor terminal arrival angle based on non-global CSI fingerprints.This method simplifies the acquisition scheme,and designs an MLP regressor to learn the closed set fingerprint feature representation and predict the CSI open set under the uncollected points.The experimental results show that it can make the open space risk about 21.67% in the angle of arrival regression prediction task,and the performance is good.Secondly,for the problem of complex environment disturbance,this thesis proposes a CSI open set recognition method based on improved Res Net and feature outlier detection.Through the improved Res Net extraction feature encoding,on this basis,the extreme value theory and outlier detection are used for CSI open set recognition,and the recognition rate is 97.89%,which is better than other open set recognition methods based on deep features.Finally,in order to solve the suboptimal graph problem of long-term maintenance of fingerprint database using graph incremental learning method,this thesis proposes a CSI open set feature incremental classification method based on Continuously Evolved Classifiers(CEC)and graph data enhancement.Create a graph structure for the fingerprint library,and optimize the fingerprint learning method by continuously expanding the CSI open set features into the fingerprint library and synchronously simplifying the graph structure of the fingerprint library.The enhanced CEC has a 3.9% improvement in the classification accuracy of the CSI open set positioning task,and the graph structure is simplified by about 80%.In summary,the application method proposed in this thesis has a certain degree of effectiveness in reducing the consumption of artificial resources and combating interference from indoor complex environments.It also has a good application prospect in the application research of indoor terminal positioning.