Research On Localization And Personal Identification Using CSI And Depth Image

Recently,indoor monitoring has received more and more attention.Personal identification and localization technologies have been widely used in various fields.However,due to the complexity of the environment,existing technology is difficult to obtain accurate identification and localization performance.In order to solve the above problem,this thesis studied localization and personal identification technique based on channel state information(CSI)and depth images.The main work includes:(1)Related theories and methods of localization and personal identification are studied.Firstly,the measurement data method and preprocessing technique are introduced.And then the software and hardware platforms of CSI data and depth image acquisition are built respectively,which provides a solid theoretical foundation for the performance analysis of the proposed algorithm.(2)A serial sequence based localization and personal identification algorithm is proposed.In the offline phase,the time domain,spatial domain and frequency domain of CSI amplitude information are used to construct the CSI amplitude image.And then the Grab Cut image segmentation algorithm is proposed to extract the target depth information in the depth image.Next,the convolutional neural network(CNN)is used for personal identification and localization training using CSI amplitude images and target depth information,respectively.In the online phase,the target depth information is used to estimate the position,and then the CSI amplitude image is used to achieve personal identification.The proposed algorithm determines the target position at first.And then the personal identification is performed next.Thus,it greatly reduces the influence of the position estimation for identification process.The experimental results demonstrate the performance of the proposed algorithm.(3)A front-end fusion and attention mechanism based localization and personal identification algorithm is proposed.In the offline phase,image pixel-level fusion is used to obtain the fingerprint image of the training data by the constructed CSI magnitude image and the segmented target depth information.Then,CNN combined with SENet attention mechanism algorithm is used for localization and identification training,respectively.In the online phase,the fused fingerprint images are used for position estimation and identification by obtained location and identity classification functions in the offline phase.In the proposed algorithm,the pixel-level image fusion can obtain more image detail information.Moreover,an attention module increases the weight of important information and reduces the weight of useless information which can significantly improve the efficiency of offline learning.The experimental results demonstrate the performance of the proposed algorithm.