基于神经网络的多传感器融合PDR定位方法

摘要

In pedestrian dead reckoning(PDR)system,natural walking motion and sensor drifts always cause the problems of inaccurate step length estimation and cumulative error in the orientation computation.This paper presents a novel NN-based multiple sensor fusion method for PDR indoor localization. Firstly,pedestrian movement estimation model is trained by BP neural network on accelerometer data,and used to predict walking distance.Secondly,a micro-heading angle fusion algorithm is proposed according to the continuity of walking and correlation of sensors.The algo-rithm classifies the walking conditions to obtain the reliable sensor source and can accurately estimate pedestrian walking direction by the weighted fusion of three kinds of micro-heading angles.The experiment results show the good performance of localization errors implemented by the micro-heading angle fusion algorithm.%针对当前行人航位推算系统因行人随意性行走、传感器漂移等造成行人步长估计不精确、方向计算误差累积问题,提出了一种基于神经网络和智能手机内置多传感器融合的PDR室内定位方法.首先利用加速计采集的传感器数据和移动距离数据训练BP神经网络,将训练好的BP神经网络模型进行行人移动距离预测,然后根据行人行走步伐的连续性特点和传感器输出之间的相关性,设计了一种微航向角融合的方向估计算法.该算法通过对行走过程中的情况进行分类以获得可靠的传感器源,利用3种微航向角进行分类加权融合,最终获得行人行走方向的精确估计.实验结果表明,通过行人移动距离预测和微航向角融合算法能够实现得较好的定位效果.