基于地磁敏感度的室内定位算法的研究

发布时间：2018-03-28 00:07

本文选题：地磁场　切入点：粒子滤波　出处：《东华理工大学》2017年硕士论文

【摘要】：现如今,人们的日常活动大多在室内进行,导致基于位置感知的应用的激增,对于位置服务的需求也与日俱增。室内定位主要存在三方面的难点:首先,障碍物,干扰源较多,室内环境呈现出较强的动态性。其次,测量误差。最后,定位往往需要依赖于其他辅助设备,设备本身的局限性。针对以上定位的三大难点,综合室内地磁场的分布特点。在无干扰源的条件下,室内地磁场变化微弱。但真实室内空间,随机分布的诸多干扰源形成地磁场的独特分布。为此本文利用干扰源对磁场特征的影响完成定位过程,首先利用特征梯度去建立特征指纹库,以此来消除辅助设备干扰。但利用地磁场作为定位载体依然面临着一大问题。复杂的室内分布也会存在局部的特征分布平滑区域,称之为“盲区”,例如,走廊,过道等区域。为解决此类问题,基于数理统计方法的粒子滤波,以总体的样本分布去估测真实位置点分布,利用粒子滤波匹配明显的特征分布区域。另外,本文通过建立预处理过程进行区域划分,待区域识别完成后,建立自适应粒子滤波,从而解决“盲区”定位问题。本文选取一个400平方米样本空间,通过移动设备采集室内空间样本点的地磁场强度和三分量值,经最临近,三次样条,双线性,双三次四种插值方法分别建立室内特征分布指纹库。之后,进行特征区域划分,并且利用观测值去寻找最佳匹配方案。最后,将感知信息融入粒子滤波去建立自适应过程,自适应优化滤波参数。此预处理过程建立测试点和指纹库之间的关联,以便于更好的去识别测试点。本文主要解决以下三个方面的问题。首先,通过有限样本点建立室内地磁特征指纹库。进而,利用室内地磁场特征分布的独特性和特征指纹库与测试点之间的关联度,对室内空间进行敏感度区域划分。最后,将室内特征分布与测试点之间的关联信息融入粒子滤波,自适应匹配环境特征,自动优化滤波参数。通过测试,最临近插值所建立的指纹库效果最差。自适应粒子滤波较传统粒子滤波算法,定位误差稳定,精度高。在不同分辨率条件下,191*191,381*381,951*951,经过100次随机定位测试,定位误差分别平均减小0.41m,1.25m,1.71m。
[Abstract]:Nowadays, most of people's daily activities are carried out indoors, which leads to the proliferation of location-aware applications and the increasing demand for location-based services. There are three main difficulties in indoor positioning: first, there are many obstacles and sources of interference. Indoor environment shows strong dynamic. Secondly, measurement error. Finally, positioning often depends on other auxiliary equipment, the limitations of the equipment itself. The distribution of indoor geomagnetic field is synthesised. The change of indoor geomagnetic field is weak under the condition of no interference source. But the real indoor space, The random distribution of many interference sources forms the unique distribution of the geomagnetic field. In this paper, the influence of the interference source on the magnetic field features is used to complete the localization process. Firstly, the characteristic fingerprint database is established by using the characteristic gradient. But using the geomagnetic field as a positioning carrier still faces a major problem. Complex indoor distributions can also have local feature distribution smoothing areas, known as "blind areas," such as corridors. In order to solve this problem, particle filter based on mathematical statistics is used to estimate the distribution of real location points with the total sample distribution, and the particle filter is used to match the obvious feature distribution area. In this paper, the preprocessing process is established to divide the area, and the adaptive particle filter is set up to solve the problem of "blind zone" location after the region recognition is completed. In this paper, a sample space of 400 square meters is selected. The geomagnetic field intensity and three component values of indoor spatial sample points were collected by mobile devices. The fingerprint database of indoor feature distribution was established by the nearest, cubic spline, bilinear and bicubic interpolation methods, and then the feature regions were divided. Finally, the perceptual information is incorporated into the particle filter to establish the adaptive process and the filter parameters are adaptively optimized. The preprocessing process establishes the correlation between the test points and the fingerprint database. In order to better identify the test points. This paper mainly solves the following three problems. Firstly, the indoor geomagnetic feature fingerprint database is built through limited sample points. By using the uniqueness of the indoor geomagnetic field characteristic distribution and the correlation degree between the characteristic fingerprint database and the test points, the sensitivity region of indoor space is divided. Finally, the correlation information between indoor characteristic distribution and test point is incorporated into particle filter. Adaptive matching of environmental features, automatic optimization of filtering parameters. By testing, the closest interpolation fingerprint database is the worst. The adaptive particle filter is more stable than the traditional particle filter algorithm, and the localization error is stable. The accuracy is high. After 100 random positioning tests, the positioning error is reduced by 0.41mt / 1.25mt / 1.71mrespectively under different resolution conditions.
【学位授予单位】：东华理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN713;P318

【相似文献】