Research On Indoor Localization Using Wi-Fi Signal And Multimodal Learning

Recently,indoor localization technology has received more and more attentions.Location based services has been widely used in various fields.However,due to the complexity of the environment,it is difficult to obtain accurate position estimation for many existing indoor localization methods.In order to solve the above problem,this thesis studied Wi-Fi signal based localization algorithm by multi-modal learning technique which makes full use of the complementarity between various modes of Wi-Fi signal.The main work includes:(1)The related theory and method of Wi-Fi signal based indoor localization is studied.Firstly,different measurement modes of Wi-Fi signal are introduced.And then the multi-modal learning theory is described in detail.Finally,the fingerprint positioning algorithm model is introduced which provides a solid theoretical foundation for the following research.(2)An amplitude and phase modalities of channel state information(CSI)measurement based localization algorithm by local preserving mapping(LPP)is proposed.Firstly,the CSI phase mode is preprocessed by linear compensation.Then the LPP method is used to extract the features of amplitude and phase modalities.Finally,support vector machine(SVM)is proposed for classification learning and obtain the position classification model.Experimental results show that the proposed algorithm performs better than the classical indoor fingerprint based localization algorithms.(3)An amplitude modality,phase modality of CSI measurement and received signal strength indication(RSSI)modality based localization algorithm by Stacking regression learning is proposed.Firstly,after the CSI phase modality preprocessing by linear compensation,the fingerprint of training data is constructed by CSI and RSSI measurements.Then the regression learning is performed using Stacking regression model.In the proposed Stacking regression model,Gaussian kernel and polynomial kernel of SVR model,Ridge model,Lasso model and Elastic-Net model are used as the base regressors for the first layer.The SVR Gaussian kernel model is used as the fusion regressors for the second layer.In the proposed algorithm,Stacking models can ensure better offline learning performance by aggregating multiple regressors.Experimental results show that the localization accuracy of the proposed algorithm is better than that of the classical fingerprint based localization algorithms.