WLAN位置指纹室内定位关键技术研究

发布时间：2018-09-09 09:00

【摘要】：移动智能终端的广泛应用和无线网络的快速普及和大量应用,使得基于位置服务(Location Based Services, LBS)的应用需求呈现出快速、大幅增长趋势,LBS迅速发展和普及到了社会生活和生产的各个领域,并逐渐显示出了良好的技术发展前景和巨大的应用市场空间。借助位置信息需求,定位技术与基于位置的服务的发展紧密地联系在一起了。其中,可靠而高效的室内定位技术是实现LBS的前提和关键所在。在已有的室内定位技术中大多需要额外的专用硬件设施,定位成本高,定位精度和覆盖范围受硬件条件的限制,不利于LBS在室内环境的应用和推广。基于无线局域网(Wireless Local Area Network, WLAN)和接收信号强度(Received Signal Strength, RSS)的室内定位技术,充分利用现有的WLAN公共基础设施,无需任何其它专用设备,只需特定的定位软件,即可通过移动智能终端实现定位。WLAN的室内定位技术因其较低的定位成本、且能满足大多数室内应用对定位精度的需求,已经成为室内定位技术的首选。但是,随着室内无线接入点的广泛部署和智能终端设备的不断增减,室内无线电传播环境越来越复杂,RSS表现出高度的多变性和复杂性,这严重影响了基于RSS的WLAN指纹定位系统的定位精度,给基于WLAN的位置指纹室内定位技术带来了全新的研究内容,也给研究工作者提出了更为艰难的挑战。本文对WLAN的基于RSS的位置指纹定位技术开展了较为深入地调查和研究。着眼实用,以LBS为研究应用背景,围绕提高RSS信号的可信度这一关键问题,以提高室内定位的可靠性和有效性为研究目标,以基础理论研究为主,采用软件与硬件结合、仿真和实验并重的研究方法,对WLAN指纹定位的定位区域聚类、AP选择、RSS信号定位特征提取等主要技术环节进行了研究。主要贡献归纳为以下几点： ·调查研究了室内RSS信号的分布特点。为了更好地描述RSS信号分布,本文选取了四种典型的室内环境(普通住宅、办公楼、教学楼和商场)进行信号收集,分析了人员、接收器方向以及样本数量对RSS信号的影响。提出了一种基于改进的双峰高斯模型(Improved Double-peak Gaussian Distribution, IDGD)定位算法。实验证明,与传统的基于直方图和高斯模型的定位技术相比,在保证相同定位精度前提下,基于IDGD的定位算法可减少大约70%的样本数量。因此IDGD算法可以大幅度地减少样本数量,减少数据采集工作量,节约定位成本,提高系统的定位精度。 ·研究了大定位目标区域的聚类问题。在较大范围的室内定位环境,RSS的统计特性变化更大,对于基于学习型定位算法来说,对整个定位区域进行学习,将增加算法复杂度,建立的定位模型不是最优的,从而不利于提高系统的定位精度。若采用聚类算法,将大的定位目标区域划分为若干个较小的定位子区域,然后在每个定位子区域建立区域定位模型,将降低计算复杂度,提高定位精度。本文针对已有的聚类分块问题没有考虑信号的相关性,从而导致分类精度不够高的问题,提出了一种将RSS信号白化后再进行k-means聚类的算法。实验表明,与k-means聚类算法相比,本文提出的聚类算法可平均提高3.7%的聚类准确度,更有利于降低系统计算复杂度,节约终端能耗,提高定位精度。 ·研究了接入点(Access Point, AP)选择问题。来自不同AP的RSS信号所包含的信息量是不同的,在当前各个公共热点高密度部署AP情况下,这种差异尤为明显。因此并不是所有的AP提供的RSS信号都有利于定位,很多RSS可能受到各种各样的噪声影响,含有大量的冗余信息,不仅不会提高系统的定位精度,反而起到相反的作用。因此需要对RSS信号也就是AP的定位能力进行判别,筛选出最优的AP集合用于定位。针对已有的AP选择算法没有考虑AP的查全率和查准率问题,本文基于信息熵理论,提出了基于信息增益权重的AP选择算法。实验表明,利用该算法优化的AP定位子集合,更有利于去掉冗余的AP,提高定位算法的解算效率和定位精度。 ·研究了提取RSS信号的有效定位特征问题。采用特征提取算法提取RSS信号的定位特征,有利于去掉RSS信号包含的冗余信息,提高RSS信号的可信度。本文针对已有算法只考虑有效提取RSS的线性特征问题,提出了基于核函数的直接判别特征提取(KD-LDA)算法,该算法可充分利用RSS信号的非线性特征。实验表明,联合本文提出的聚类和AP选择算法,采用学习机器支持向量回归定位模型,可实现在1米内的定位精度置信概率达37.1%,最大误差为4.12米。与传统的定位算法相比,可显著提高系统小误差定位(1米内)的概率,缩小系统的定位误差范围,优化系统定位性能。
[Abstract]:With the wide application of mobile intelligent terminal and the rapid popularization of wireless network, the application demand based on Location Based Services (LBS) has shown a rapid and substantial growth trend. LBS has rapidly developed and popularized to all fields of social life and production, and has gradually shown a good technical development prospects. Location technology is closely related to the development of location-based services with the help of location information requirements. Reliable and efficient indoor location technology is the premise and key to LBS.
Most of the existing indoor positioning technologies require additional dedicated hardware facilities. The cost of positioning is high. The positioning accuracy and coverage are limited by hardware conditions, which is not conducive to the application and promotion of LBS in indoor environment. Indoor positioning technology, which makes full use of existing WLAN public infrastructure, does not need any other special equipment, only needs specific positioning software to achieve positioning through mobile intelligent terminals. Because of its low cost of positioning, and can meet the needs of most indoor applications for positioning accuracy, WLAN indoor positioning technology has become indoor positioning. However, with the widespread deployment of indoor wireless access points and the continuous decrease of intelligent terminal equipment, indoor radio propagation environment is becoming more and more complex, RSS shows a high degree of variability and complexity, which seriously affects the positioning accuracy of WLAN fingerprint positioning system based on RSS, and gives indoor positioning of location fingerprint based on WLAN. Technology brings a whole new research content, and also poses a more difficult challenge for researchers.
In this paper, the location fingerprint positioning technology based on RSS in WLAN is investigated and studied in depth. With the practical point of view and LBS as the application background, the key problem of improving the reliability and validity of RSS signal is focused on to improve the reliability and validity of indoor positioning. Combined with simulation and experiment, the main technical links of WLAN fingerprint location, such as location region clustering, AP selection, RSS signal location feature extraction, are studied.
In order to describe the distribution of RSS signals better, this paper selects four typical indoor environments (ordinary residential buildings, office buildings, teaching buildings and shopping malls) for signal collection, and analyzes the effects of personnel, receiver direction and sample size on RSS signals. Experimental results show that IDGD algorithm can reduce the number of samples by about 70% compared with traditional localization techniques based on histogram and Gaussian model under the same localization accuracy. It can reduce data collection workload, save location cost and improve positioning accuracy of the system.
The clustering problem of large localization target area is studied. In a wide range of indoor localization environment, the statistical characteristics of RSS change much more. For learning localization algorithm, learning the entire localization area will increase the algorithm complexity, and the localization model is not optimal, which is not conducive to improving the positioning accuracy of the system. Clustering algorithm is used to divide the large localization target region into several smaller localization sub-regions, and then a regional localization model is established in each localization sub-region, which will reduce the computational complexity and improve the localization accuracy. The experimental results show that compared with K-means clustering algorithm, the clustering algorithm proposed in this paper can improve the clustering accuracy by an average of 3.7%, which is more conducive to reducing the system computational complexity, saving terminal energy consumption and improving the positioning accuracy.
Access Point (AP) selection is studied. RSS signals from different APs contain different amounts of information, especially in the current situation of high density deployment of AP in public hot spots. Because there is a lot of redundant information, it will not improve the positioning accuracy of the system, but will play the opposite role. Therefore, it is necessary to discriminate the positioning ability of RSS signal, that is, AP, and select the best AP set for positioning. In theory, an AP selection algorithm based on information gain weight is proposed. Experiments show that the optimized AP subset is more conducive to remove redundant AP and improve the algorithm efficiency and positioning accuracy.
This paper studies the problem of extracting the effective location features of RSS signals.Using the feature extraction algorithm to extract the location features of RSS signals is helpful to remove the redundant information contained in RSS signals and improve the reliability of RSS signals.Aiming at the problem that the existing algorithms only consider the effective extraction of the linear features of RSS signals,a direct discriminant feature based on kernel function is proposed. Experimental results show that the proposed clustering and AP selection algorithm combined with the learning machine support vector regression localization model can achieve a confidence probability of 37.1% and a maximum error of 4.12 meters within 1 meter. Compared with the traditional localization algorithm, the proposed algorithm can be significantly improved. Increase the probability of small error positioning (within 1 meter), reduce the range of positioning error, and optimize the positioning performance of the system.
【学位授予单位】：华东师范大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：TN925.93

【参考文献】