测绘学报 ›› 2015, Vol. 44 ›› Issue (12): 1322-1330.doi: 10.11947/j.AGCS.2015.20130780

• 大地测量学与导航 • 上一篇    下一篇

面向室内WLAN定位的动态自适应模型

吴东金1,2, 夏林元1,2   

  1. 1. 中山大学地理科学与规划学院, 广东 广州 510275;
    2. 中山大学广东省城市化与地理环境空间模拟重点实验室, 广东 广州 510275
  • 收稿日期:2013-12-18 修回日期:2015-06-30 出版日期:2015-12-20 发布日期:2016-01-04
  • 通讯作者: 夏林元,E-mail:xialiny@mail.sysu.edu.cn E-mail:xialiny@mail.sysu.edu.cn
  • 作者简介:吴东金(1985-),男,博士生,研究方向为无缝定位、室内定位与LBS。E-mail:wudj-no.15@163.com
  • 基金资助:
    国家自然科学基金(41071284);广东省重大科技专项(2015B010104003)

Dynamic Adaptive Model for Indoor WLAN Localization

WU Dongjin1,2, XIA Linyuan1,2   

  1. 1. School of Geography and Planning, Sun Yat-sen University, Guangzhou 510275, China;
    2. Guangdong Key Laboratory for Urbanization and Geo-simulation, Sun Yat-sen University, Guangzhou, 510275, China
  • Received:2013-12-18 Revised:2015-06-30 Online:2015-12-20 Published:2016-01-04
  • Supported by:
    The National Natural Science Foundation of China (No. 41071284);Science and Technology Planning Projects of Guangdong Province (No.2015B010104003)

摘要: 提出了一种面向无线局域网(wireless local area network,WLAN)位置指纹匹配定位的动态自适应模型,借助多个基站的实时数据为稳健室内定位提供更新的匹配数据库——radio map。基于基站与radio map采样点之间的空间关联性,将基站数据和radio map分别作为多层神经网络的输入和输出,动态更新radio map;利用多元数据异常探测技术检验基站数据捕获环境的时空变化;根据探测结果采用顾及室内布局的数值内插和再训练的方式更新模型,从而使其适应环境的时空变化。在室内动态环境中进行了多次验证试验。试验结果显示,在时变因素作用下,相比较传统方法,采用所提模型的定位方法的平均误差至少下降10%;在空间变化因素(以信标移动为例)作用下,其他方法平均定位误差大幅增加了至少165%,而采用所提模型方法的平均定位误差只增加了10%~20%,定位精度维持在3 m左右(即原始精度)。结果证明采用了所提模型的定位系统能够自适应环境的时空变化而保持原有定位表现。不过,与传统位置指纹匹配定位方法相比,该模型带来了更多的计算负荷。

关键词: 室内定位, 无线局域网, 动态自适应模型, 神经网络, 多元异常值检测

Abstract: To support robust indoor localization, it is presented that a dynamic adaptive model (DAM) for WLAN (wireless local area network) location fingerprinting which can provide updated radio maps depending on the real time data from several base stations (BS). The model takes the spatial relationships between the BSs and the sample points of the radio map into account that the data of BSs and radio map is respectively used as the inputs and outputs of multilayer neural networks to update radio maps dynamically. In order to catch tempo-spatial environmental changes, the multivariate outlier detection technique is applied to examine the data of BSs. According to the detecting results, a retraining process and an interpolation method considering the floor plan are used to update the functional model and make the model adapt to tempo-spatial environmental changes. The model is evaluated in indoor dynamic environments. Compared to conventional ones, the average location error of the proposed model-based method decreases more than 10% in time-varying environments; and after spatial environmental changes (radio beacons are moved), its average location error increases 10% to 20% which is much lower than 165% increase of others. Moreover, the localization accuracy is around 3 m, holding the original performance. The results prove the adaptation of the proposed model to the tempo-spatial environmental changes. However, compared to conventional location fingerprinting, the model brings a little more computational overhead.

Key words: indoor localization, WLAN, dynamic adaptive model, neural networks, multivariate outlier detection

中图分类号: