Research On Indoor Localization Technology Using Channel State Information And Depth Image

With the continuous development of society,location based service has been widely used in various fields.However,a single source based indoor localization technology often cannot meet high-precision location requirement.Therefore,in order to improve the indoor localization performance,this thesis studied channel state information(CSI)and depth image based localization algorithm by machine learning and multi-source fusion theory.The main work of the paper is described as follows:1.The basic theory of indoor localization is studied.First,the existing CSI and image localization technologies and methods are introduced.And then the multi-source information fusion strategy is described.Finally,an experimental platform is built to provide the foundation for the following research.2.A CSI and depth image indoor localization algorithm by decision-level fusion is proposed.First,the CSI amplitude information is converted into CSI images,the depth images are preprocessed by image segmentation.Then the convolutional neural network(CNN)is used for location classification learning by CSI image and depth image,respectively.Two localization classification models are obtained.Finally,the decision-level data fusion strategy is used to obtain final position estimation result.In the proposed algorithm,decision-level data fusion has strong fault tolerance and can improve the performance of classification learning.The experimental results show that the proposed algorithm can obtain better positioning results.3.A CSI and depth image indoor localization algorithm by feature-level fusion and attention mechanism is proposed.Firstly,after preprocessing CSI amplitude information and depth image,CNN is used to extract feature information,and then the feature-level fusion strategy is used for feature fusion.Finally,the CNN network combined with the attention mechanism is used for position regression learning.In the proposed algorithm,feature-level fusion can fully represent measurement features.The attention mechanism can focus on important information.Both of these two process can improve the localization performance dramatically.The experimental results verify the localization performance of the proposed algorithm.