节点文献
基于室内多信息辅助的行人航迹推算技术研究
Research on Pedestrian Dead Reckoning Based on Indoor Multi-information Assistance
【作者】 陈建；
【作者基本信息】 厦门大学 ， 通信与信息系统， 2020， 博士
【摘要】 随着导航与位置服务产业的蓬勃发展,室内行人导航定位得到了广泛的关注。如何提高室内行人定位精度和鲁棒性面临着诸多挑战。目前主流的技术方案是采用磁力计、惯性传感器以及WiFi传感器进行定位。传统的磁场指纹匹配、惯导定位、多信息融合存在着指纹模糊、方向漂移引起的累积误差、多信源融合定位精度低等问题。针对这些问题,本文对室内多信息辅助的行人航迹推算技术进行了研究,研究的目的是通过合理的使用微机电惯性传感器、磁力计、WiFi传感器提供一种高精度、高可靠性、低成本的室内行人定位解决方案,研究内容包括以下三个部分:(1)针对室内磁场波动引起的指纹模糊性问题,引入惯导信息,将其建模为双模序列匹配模型,提出三维动态时间规整定位算法。该算法使用惯导姿态角信息提取出磁场指纹的垂直分量和水平分量,形成二维磁场指纹,扩展了磁场指纹的维度信息,减少磁场指纹模糊性引起的误匹配,提高磁场/惯导定位算法的定位精度和鲁棒性。实验表明使用Nexus 5构建指纹库时,教学楼、自习室、办公楼的平均定位误差分别为1.53米、1.66米、3.42米;使用Redmi Note 7构建指纹库时,教学楼、自习室、办公楼的平均定位误差分别为1.38米、1.43米、2.8米;使用Samsung A5构建指纹库时,Atlantis le Centre购物中心平均定位误差为3.78米。(2)针对惯导系统方向漂移引起的累积误差问题,结合建筑物楼层平面图提出智能粒子滤波算法,可有效解决惯导系统方向漂移引起的累积误差问题。该算法利用建筑物地图信息约束粒子的活动范围,使用萤火虫算法让无效粒子向有效粒子迁移,消除无效粒子参与行人状态估计,提高智能粒子滤波算法中有效粒子多样性,从而降低室内行人定位误差。实验表明教学楼、自习室、办公楼、Atlantis le Centre购物中心的平均定位误差分别为1.64米、1.06米、1.28米、4.16米。(3)针对单一传感器存在系统误差,多信源融合算法存在定位精度低等问题,基于加速度计、陀螺仪、磁力计以及WiFi传感器提出增强型Kalman滤波的室内多信源融合定位算法。该算法首先使用粗差剔除机制移除磁场、WiFi指纹匹配后较大位置误差,其次使用加权矩阵、自适应观测噪声提高增强型Kalman滤波性能,减少多信源融合定位算法定位误差。实验表明使用Nexus 5构建指纹库时,教学楼、自习室、办公楼的平均定位误差分别为0.99米、1.02米、1.53米;使用Redmi Note 7构建指纹库时,教学楼、自习室、办公楼的平均定位误差分别为1.16米、1.39米、1.37米;使用Samsung A5构建指纹库时,Atlantis le Centre购物中心平均定位误差为2.5米。最后,对本论文研究的室内多信息辅助的行人航迹推算算法进行了总结,并对未来工作进行了展望。
【Abstract】 With the vigorous development of navigation and location service industry,indoor pedestrian navigation and positioning has received widespread attention.How to improve the accuracy and robustness of indoor pedestrian positioning is faced with many challenges.At present,the mainstream technical scheme is to use magnetometer,inertial sensor and WiFi sensor for positioning.There are some problems in traditional magnetic field fingerprint matching,inertial navigation positioning and multi-information fusion,such as fingerprint Fuzziness,cumulative error caused by direction drift,low accuracy of multisource fusion positioning and so on.To solve these problems,this paper studies the indoor multi-information assisted pedestrian dead reckoning technology.The purpose of the research is to provide a high-precision,high-reliability and low-cost indoor pedestrian positioning solution through the rational use of MEMS inertial sensor,magnetometer and WiFi sensor.The research content includes the following three parts:(1)Aiming at the problem of fingerprint fuzziness caused by indoor magnetic field fluctuation,based on inertial navigation,the magnetic field matching is modeled as a two-mode sequence matching model,and a three-dimensional dynamic time warping localization algorithm is proposed.The algorithm uses the inertial navigation attitude angle information to extract the vertical and horizontal components of the magnetic field fingerprint to form a two-dimensional magnetic field fingerprint,which expands the dimensional information of the magnetic field fingerprint and reduces the mismatching caused by the fuzziness of the magnetic field fingerprint.Finally,the positioning accuracy and robustness of the magnetic field/inertial navigation positioning algorithm are improved.The experimental results show that when using Nexus 5 to build fingerprint database,the average positioning errors of teaching building,study room and office building are 1.53 m,1.66 m and 3.42 m respectively.When using Redmi Note 7 to build fingerprint database,the average positioning errors of teaching building,study room and office building are 1.38 m,1.43 m and 2.8 m respectively.When using Samsung A5 to build fingerprint database,the average positioning error of Atlantis le Centre shopping center is 3.78 m.(2)Aiming at the problem of cumulative error caused by the direction drift of inertial navigation system,combined with building floor plan,an intelligent particle filter algorithm is proposed,which can effectively solve the problem of cumulative error in inertial navigation system.In this algorithm,the activity range of particles is constrained by the information of building map,and the firefly algorithm is used to make invalid particles migrate to effective particles,so as to eliminate invalid particles to participate in pedestrian state estimation and improve the effective particle diversity of intelligent particle filter algorithm.Finally,the indoor pedestrian positioning error is reduced.The experimental results show that the average positioning errors of teaching building,study room,office building and Atlantis le Centre shopping center are 1.64 m,1.06 m,1.28 m and 4.16 m respectively.(3)In order to solve the problems of system error of single sensor and low positioning accuracy of multi-source fusion algorithm,an indoor multi-source fusion positioning algorithm based on enhanced Kalman filter is proposed based on accelerometer,gyroscope,magnetometer and WiFi sensor.Firstly,the gross error elimination mechanism is used to remove the large position error in the magnetic field and WiFi fingerprint matching estimation,and then the weighted matrix and adaptive observation noise are used to improve the performance of the enhanced Kalman filter and reduce the position error of the multisource fusion algorithm.The experimental results show that when using Nexus 5 to build fingerprint database,the average positioning errors of teaching building,study room and office building are 0.99 m,1.02 m and 1.53 m respectively.When using Redmi Note 7 to build fingerprint database,the average positioning errors of teaching building,study room and office building are 1.16 m,1.39 m and 1.37 m respectively.When using Samsung A5 to build fingerprint database,the average positioning error of Atlantis le Centre shopping center is 2.5 m.Finally,the indoor multi-information-aided pedestrian dead reckoning studied in this paper is summarized,and the future work is prospected.