基于粒子滤波的WiFi行人航位推算融合室内定位

发布时间：2018-04-11 15:25

本文选题：室内定位 + 多传感器融合　；参考：《计算机应用》2016年05期

【摘要】：为提高室内定位的精度和稳定性,提出使用粒子滤波融合WiFi指纹定位和行人航位推算的室内定位算法。为减少复杂室内环境对WiFi指纹定位的影响,提出将支持向量机分类与回归相结合的两级WiFi指纹定位算法。在基于智能手持设备惯性传感器的行人航位推算中,为减少惯性传感器的误差以及人随意行走带来的影响,采用状态转换的方法识别行走周期并进行步数统计,提出根据实时加速度数据动态设置状态转换的参数,利用步长和垂直加速度之间的关系以及相邻步长之间的关系,应用卡尔曼滤波进行步长计算。仿真实验中,基于支持向量机的WiFi指纹定位的平均误差比最近邻居(NN)算法降低34.4%,比K最近邻居(KNN)算法降低27.7%。改进的行人航位推算的性能优于常用代表性计步软件和步长计算算法,而经过粒子滤波融合后估计的行走轨迹更加接近真实轨迹:直线行走平均误差为1.21 m,优于WiFi的3.18 m和航位推算的2.76 m;曲线行走平均误差为2.75 m,优于WiFi的3.77 m和航位推算的2.87 m。
[Abstract]:In order to improve the accuracy and stability of indoor positioning, an indoor localization algorithm using particle filter fusion of WiFi fingerprint location and pedestrian carrier reckoning is proposed.In order to reduce the influence of complex indoor environment on WiFi fingerprint location, a two-level WiFi fingerprint location algorithm combining support vector machine classification with regression is proposed.In order to reduce the error of inertial sensor and the influence of human walking at will, the method of state transition is used to identify the walking cycle and count the steps in the calculation of pedestrian position based on the inertial sensor of intelligent handheld device.According to the real time acceleration data, the parameters of state conversion are dynamically set, and the Kalman filter is used to calculate the step size by using the relation between step size and vertical acceleration and the relation between adjacent step sizes.In the simulation experiment, the average error of WiFi fingerprint location based on support vector machine is 34.4 lower than that of nearest neighbor (NN) algorithm, and 27.7m lower than that of K-nearest neighbor (KNN) algorithm.The performance of improved pedestrian carriageway calculation is better than that of typical step software and step size calculation algorithm.After particle filter fusion, the estimated trajectory is closer to the real track: the average error of straight line walking is 1.21 m, which is better than that of WiFi (3.18 m) and WiFi (2.76 m), the average error of curve walking is 2.75 m, which is better than that of WiFi (3.77 m) and WiFi (2.87 m).
【作者单位】：电子科技大学信息与软件工程学院;
【基金】：国家科技支撑计划项目(2012BAH44F00)~~
【分类号】：TN92;TP18

【相似文献】