Research On Indoor 3D Localization Algorithm Of Wi-Fi Based On Multi-Classifier Fusion

Indoor localization technology based on received signal strength fingerprinting is widely used in daily life and industry.Especially during the COVID-19 pandemic,it provides technical support for contactless work and life.After years of research and development of localization algorithms,many innovative solutions have been proposed for indoor localization.The multi-classifier fusion-based localization technology has achieved a certain level of improvement in terms of localization accuracy,when compared to traditional localization methods.It is a hot topic in indoor localization technology research.However,there are still certain limitations.(1)In indoor localization,the complex indoor environment can cause multipath effects in the propagation of Wi-Fi signals,resulting in localization deviations in the localization process.(2)In multi-floor localization,existing fusion localization technologies ignore the mutual influence between different floors during the localization process,resulting in low indoor localization accuracy.(3)Traditional fusion localization technologies require a huge amount of computing resources,resulting in low efficiency during indoor localization.By improving the fusion algorithm,we can further improve both the localization accuracy and efficiency.This paper proposes two fusion-based indoor localization algorithms from different fusion perspectives.(1)The first algorithm aims at the problem of false localization between floors that the existing fusion localization technology ignores in multi-floor localization and localization errors caused by multipath effects.A Wi-Fi indoor 3D localization algorithm with multiple classifiers fusion,called FLMCF,is proposed.Firstly,floor classification training is carried out to reduce localization deviation in the vertical direction.Secondly,in this algorithm,multiple classifiers are used to train models for each floor.Additionally,the optimal set of weights is trained using the method of average minimum localization error.This algorithm can fully integrate the advantages of each classifier in order to reduce the influence of multipath effects and improve localization accuracy.Finally,the Reliability Fusion Weight Selection(RFWS)algorithm is used to select the weight with high confidence level,and calculate the final location coordinates based on this.The experimental results showed that FLMCF improved both the localization accuracy and stability compared to traditional machine learning localization algorithms and existing fusion localization algorithms.Additionally,it increased the localization timeliness while ensuring the localization accuracy.(2)The second algorithm is aimed at three issues.In the process of FLMCF fusion,limited by the initial weights,the training fusion weight set can only be relatively optimal.The fusion localization algorithm requires a large amount of computing resources,which affects the localization efficiency.The inherent multipath effect of the complex indoor environment affects the localization accuracy.A neural network-based multi-classifier fusion Wi-Fi indoor 3D localization algorithm is proposed,called MCREFLoc.First,perform floor classification training and use multiple classifiers to train the regional localization model on each floor,so as to reduce the incorrect localization between floors and the influence of multipath effects.Next,a regression model based on neural network is trained to automatically adjust the weights of each classifier and optimize the fusion process.Finally,the regression model based on the neural network is used to fuse the initial prediction results of multiple classifiers to calculate the final location coordinates.By fully utilizing the generalization ability of the neural network,the proposed method avoids the limitations of the initial weights and expands the number of weights,leading to improved performance in indoor localization.The experimental results showed that the MCREFLoc algorithm reduced localization error compared to traditional localization classifiers,classical convolutional neural networks,and existing fusion localization algorithms.Moreover,it relatively improved localization timeliness and stability.