| The rapid development of the Internet of Things technology,and the increasing demand for indoor positioning services in many fields such as life,commerce,and military,have stimulated the rapid development of indoor positioning technology.At present,the research on indoor positioning technology has made progress in certain stages.Emerging positioning technologies such as visible light positioning are becoming more and more perfect.WiFi and Bluetooth positioning technologies have been successfully used in commercial and public services.UWB and visual positioning technologies are gradually meeting the needs of high-precision industries.With the continuous reform and advancement of technology,the development of positioning technology has become increasingly mature.Data characteristics of the positioning signal,the applicable range of the positioning algorithm and the performance characteristics of the positioning system have been quite clear.Based on the current development trend,it is not only necessary to further optimize the existing positioning technology to improve positioning accuracy,but also to efficiently integrate the existing positioning technology to achieve a better positioning effect.The fusion positioning under the hybrid network can fully improve the accuracy and robustness of the positioning system by fully tapping the complementary advantages of different positioning systems,positioning features,and positioning algorithms.This article focuses on indoor fusion positioning under a hybrid network,and does the following:(1)This paper proposes a unified architecture for a hybrid construction fusion positioning system,and examines the determinants of fusion positioning accuracy from three perspectives: positioning source,algorithm,and weight space.In addition,this article summarizes the feature fusion,decision fusion,and multi-level fusion problems based on the fused information.(2)Aiming at the problem of feature fusion,this paper studies a feature fusion algorithm based on deep networks.The algorithm uses deep network learning to locate fingerprint deep features,non-linear feature mapping networks unify feature dimensions,and feature stitching to generate new features containing hybrid feature information.The measured data results show that constructing new localization features can effectively excavate the advantages of mixed features,strengthen the correlation between different features,and achieve complementary hybrid features.(3)In order to improve the estimation accuracy of positioning decisions,this paper proposes a candidate set-assisted two-layer fusion network algorithm.This algorithm constructs various layers and fusion layers.The former generates multiple positioning decisions by combining different features and algorithms,and the latter calculates the fusion contour to measure the accuracy of different positioning modes,and weights the fusion to get the final position estimate.Among the various layers,the proposed algorithm measures the performance of the positioning mode through a convergence function,achieves the maximum probability including correct positioning decisions,and constructs a candidate set to increase the effective information required for fusion positioning.Based on actual data verification,the proposed decision fusion algorithm achieves higher positioning accuracy than other classic fusion algorithms. |
