Research On Adaptive Fusion Indoor Positioning Method Based On Channel State Information And Image Robust Features

With the process of social digitalization,the status of location-based information services in people’s lives is gradually improving,and positioning algorithms play a crucial role in it.In the indoor environment,many positioning methods have been proposed,among which images and Wi-Fi signals have become the mainstream positioning signals due to their low cost.However,the indoor scene is complex,the accuracy of the positioning method based on a single signal is limited,and it is difficult to meet the stable high-precision positioning requirements in the indoor scene.It is urgent to develop a high-precision positioning algorithm based on multi-source signal fusion.Based on this,this paper studies the fusion positioning algorithm for the channel state information(CSI)and image signals in the Wi-Fi signal.In order to extract the location features of the two kinds of signals,a joint characterization method MHS A-EC based on multi-head self-attention and effective CSI and an image domain adaptive characterization network ALSC against illumination changes are proposed respectively.In order to reduce the interference of low-texture images on the fusion representation,increase the weight of the effective signal in the fusion feature,and improve the accuracy of fusion positioning,this paper proposes an opposite orientation image mix-up strategy and an adaptive weight fusion positioning algorithm.The specific research work is as follows:Aiming at the problem of insufficient CSI feature extraction ability and long-distance point mismatch in CSI-based localization algorithms,a joint characterization method based on multi-head self-attention and effective CSI is proposed on the basis of full analysis of existing feature extraction algorithms.MHS A-EC.The algorithm uses the multi-head selfattention mechanism to aggregate non-adjacent features,extracts CSI features,and enhances the discrimination of CSI features;and uses the nonlinear relationship between effective CSI and distance,and introduces it as distance constraint information;Jointly form the characterization of the CSI.This characterization method increases the discrimination of CSI features while introducing distance constraint information.Aiming at the problem that multi-directional image representation is easy to introduce low-texture invalid signals,and fusion representation is difficult to enhance effective signals,an opposite-directional image mixup strategy and adaptive weight fusion positioning algorithm AWF-Loc are proposed to reduce the impact of low-texture image features on fused images.The interference of features and the adaptive enhancement of the effective features of different signals.The image mix-up strategy of the opposite orientation first evaluates the image texture degree,and introduces the evaluation results in the form of weights when the opposite orientation image features are fused to obtain the image features of the coordinate axis;,evaluate the contribution of features through the gating network,generate gating weights corresponding to different signals,and add the multi-source signal features and the corresponding weights to obtain fusion features,which are used for position judgment of different coordinate axes.The positioning weight of the effective signal improves the accuracy of the positioning algorithm.In the experiment to verify the positioning effect of the CSI joint representation,the positioning error of the MHSA-EC algorithm in the comprehensive office scene is 0.71m,which is 40%higher than that of the Confi algorithm based on the convolutional network.In the experiment to verify the anti-illumination change performance of the image representation,the ALSC algorithm can achieve an average error of 0.86m in a small laboratory scene after changing lighting conditions,and the positioning accuracy is improved by 10.4%compared to the positioning method of VGG-16;after obtaining the CSI joint characterization and robust image characterization,In this paper,an adaptive weight fusion localization algorithm experiment is carried out on the two representations.The average error of the adaptive weight fusion positioning algorithm in the comprehensive office scene is 0.63m,which is 11%lower than that of the MHSA-EC algorithm,and 13%lower than that of the MDS-KNN positioning method.Therefore,the joint characterization of CSI improves the feature discrimination of CSI,and the image features that are resistant to illumination changes improve the robustness of image features.The adaptive weight fusion algorithm based on the two can further enhance the effective signal and improve the positioning accuracy.