【作者】 李楠；

【导师】 陈家斌；

【作者基本信息】 北京理工大学 ， 导航、制导与控制， 2017， 博士

【摘要】 随着移动通信和定位技术的不断发展,基于WLAN的室内定位技术受到越来越多的关注,并在行人导航、紧急救援、医疗保健等领域展现出广泛的应用前景。但是,现有的WLAN室内定位技术在定位精度、实时性和灵活性上还不能完全满足人们的使用需求,限制了该技术的大规模应用和推广。通过对室内定位技术国内外研究现状的综述和分析,目前WLAN室内定位技术主要存在以下几个方面的问题:首先,受到室内复杂环境中多径效应、人员吸收、信号干扰等的影响,WLAN信号强度易出现波动和跳变,严重影响了定位的精度和稳定性;其次,现有的匹配型定位算法计算量大,在移动终端运行时易产生定位时延;最后,现有研究主要集中在平面的二维定位,对人们最为关心的楼层定位问题涉及较少。本文对WLAN室内定位技术进行了较为深入、系统的研究,针对定位技术中存在的问题,提出了相应的解决方案。本文的主要研究内容如下:为了提高定位系统的定位速度,提出了一种基于重要性加权密度函数改进模糊聚类的WLAN信号强度指纹库聚类算法。在WLAN定位中加入聚类步骤,可以有效降低定位阶段的计算量,同时还可以减少由于误匹配造成的定位误差,提高定位精度。模糊C均值聚类算法使用模糊隶属度来描述聚类结果,降低了聚类误匹配发生的概率,同时还降低了误匹配造成的定位误差。对于模糊C均值聚类算法,聚类中心的初始值选择十分重要,不同的初始值会导致不同的聚类划分结果。为了提高聚类效果,减少聚类算法的计算时间,本文提出使用重要性加权密度函数来计算初始聚类中心。重要性加权密度函数算法考虑了不同数据间的重要性和密度差异,选取了重要性权值较高且处于密集区域的数据点作为初始聚类中心,避免了聚类算法陷入奇异值和局部最优,是一种比较合理的聚类初始化方法。为了提高WLAN算法的定位精度,提出了一种基于协同粒子群优化神经网络的定位算法。该算法利用离线采集的信号强度数据,通过协同粒子群优化算法同时对神经网络的结构和参数进行训练。定位时,只需将实时接收到的信号强度输入已训练好的神经网络模型,即可获得当前的位置估计。相比于传统的指纹匹配算法,该算法在线定位阶段计算简单,定位反应迅速,且由于神经网络对非线性数据具有较强的泛化能力,因此该算法具有较高的定位精度。针对WLAN定位结果发生易发生跳变,致使误差偏大的问题,提出了一种基于自适应粒子滤波的WLAN-PDR组合定位算法。该算法将WLAN定位和行人PDR定位组合起来,通过两种定位方法的优势互补,以获得连续、稳定、准确的定位信息。同时,本文针对智能手机的硬件特征,从步态检测、步长估计、航向校准三个方面对行人PDR算法进行了改进,使其在应用在手机中进行定位时能够获得更高的定位精度。为了解决行人楼层定位的问题,提出了一种行人楼层识别方法。该方法根据行人楼层变化时产生的气压值变化对楼层变化进行检测,使用FCM聚类结果对具体楼层进行识别。通过对FCM聚类中隶属度的计算增强了识别算法的抗干扰能力,即使在WLAN信号特征不明显的区域,也能达到较好的楼层识别精度。更多还原

【Abstract】 With the development of mobile communication and location technology,WLAN based indoor location has attracted more and more attention,and shown great energy in pedestrian navigation,emergency and medical care areas.However,existing WLAN based indoor localizationcannot satisfy people’s needs in location accuracy,real time and flexibility performance,whick limits its large scale application and popularization.Through the review and analysis in the research status of indoor localization technology,the existing WLAN based indoor localization have following serval problems: firstly,with affected by multipath effect,people absorption and signal interference in complex indoor environments,WLAN signal strength suffers from fluctuation and jump,which seriously affects the location accuracy and stability.Secondly,the existing matching based location algorithm has a large calculation quantity,which easily lead to loacation delay when running in mobile terminal.Lastly,the existing researchs mainly focus on two dimensions positioning,and seldom involve in multi floor positioning problem.The thesis makes a deep and systematic study in WLAN based indoor localization.Aiming at existing problems in location technology,the thsis proposes corresponding solutions.The research contents are as follows:Proposing an Importance Weighted Density Function(IWDF)based fuzzy clustering algorithm to cluster the WLAN signal strength fingerprints databased,which helps accelerate location speed.With the helps of clustering procedure,the amount of calculation in location phase will be reduced,meanwhile it also helps reduce the location error caused by mismatching and improve the location accuracy.The Fuzzy C Means(FCM)clustering algorithm uses the fuzzy membership to describe clustering results,which helps reduce the probability of mismatching and the location error caused by mismatching.The choice of initial cluster center is important for FCM clustering algorithm,and diffierent initial cluster center will lead to different clustering results.The thesis uses IWDF to calculate the initial cluster center,which helps improve the clustering results and reduce the time consumption of clustering.As a reasonable clutering initial algorithm,the IWDF considers the difference of importance and density among location data,the data with higher importance weights and dense is chosen to be the initial cluster center,which avoids the clustering results fall into singular values and local optima.In order to improve the location accuracy of WLAN localization algorithm,the thesis proposes a Cooperative Particle Swarm Optimization(CPSO)based Artifical Neural Network(ANN).With the signal strength data collect in offline phase,the CPSO algorithm trains the structure and parameter of ANN at the same time.During the online location phase,the location results will be estimated once the online collected signal strength data is inputing to trained ANN model.Compared to the traditional fingerprinting algorithm,the CPSO algorithm has less computional works during the online location phase,and has a fast location speed.Meanwhile,due to the nonlinear generalization ability of ANN,the CPSO-ANN localization system has a high location accuracy.Aiming at the large location error caused by WLAN localization jumping,the thesis proposes an Adaptive Particle Filter(APF)based WLAN-Pedestrain Dead Reckoning(PDR)combined localization algorithm.The WLAN-PDR localization algorithm combines the advantages of two respective location algorithm,to get a continuous,stable and accurate location results.Meanwhile,aiming at improving the location accuracy in mobilephone,the thesis improves the PDR algorithm in gait detection,step length estimation and heading calibration.In order to solve the multi floor location problems,the thesis proposes a pedestrian multi floor recognition method.The method detects the floor change by measuring the change of environment air pressure,and uses the FCM clustering results to helping determine the specify floor number.The method ability of resisting disturbance is improved by using the fuzzy membership.Even in areas with little WLAN signal features,the method can achieve a well floor recognition accuracy.更多还原