【作者】 陈伟；

【导师】 王建宇； 傅忠谦；

【作者基本信息】 中国科学技术大学 ， 电路与系统， 2010， 博士

【摘要】 随着导航定位功能在手机中的普及和基于位置服务LBS(Location-Based Service)的蓬勃发展,对行人导航定位技术准确性、可靠性和连续性的要求日益迫切。然而,作为目前行人导航领域的重要定位手段,美国全球定位系统GPS(Global Positioning System)在城市峡谷和室内等复杂环境中因为信号的衰减、干扰和遮挡等,接收机定位精度很差甚至无法定位,而此类环境恰好是个人用户的主要活动区域。为了获得连续的导航定位结果,必须借助于其它定位增强技术,如自包含传感器、无线局域网WLAN(Wireless Local Area Network)、移动蜂窝网络、射频识别RFID(Radio Frequency IDentification)、伪卫星等。基于GPS和自包含传感器的导航定位技术,以其自主导航、不需要额外设施的优点,受到导航工业界和学术界的广泛关注和集中研究。本博士课题基于一套低成本多传感器定位平台MSP(Multi-Sensor Positioning,包含一个GPS接收机、一个两轴数字罗盘和一个三轴加速度计),研究在GPS接收机无法提供准确连续定位信息的情况下,如何通过自包含传感器测量行人的速度和航向,并设计组合滤波器融合GPS和自包含传感器的定位结果,实现行人室内外无缝定位解决方案。通过对各种行人无缝定位算法的调研和深入了解,考虑到硬件系统低成本传感器的性能无法满足传统惯性导航机制的精度要求,本文选择了行人航迹推算算法PDR(Pedestrian Dead Reckoning)辅助在GPS性能不佳的复杂环境中的定位,重点研究了以下三个方面:1.行人步频探测和步长估计算法研究利用行人步态的运动生理学特性,参照传统步频探测和步长估计方法,设计了基于加速度信号的集成了滑动窗口( Sliding-Window )、零交叉(Zero-Crossing)和峰值探测(Peak Detection)三种方法的步频探测算法,实现了只有一个参数的步长估计模型,通过实验评估了步频探测算法正确率和步长估计模型性能。此外,本文创新性地将肌电信号EMG(Electromyography)用于步频探测和步长估计,详细说明了行人肌肉群选择、信号预处理、探测算法实现和估计模型设计等问题,实验验证了该方法的可行性。2.数字罗盘航向校准算法的研究获取精准的航向是PDR算法的核心问题,也是困扰个人定位研究的长期难题。针对两轴数字罗盘无法实现倾斜补偿,容易受到软硬铁效应、固定偏置、比例因子和行走时平台振动等误差影响的弱点,本文采用仿真手段研究了各种因素对航向影响的误差特性,充分考虑到行人导航定位对校准方法的易操作性要求,提出了包含所有可预测误差的统一误差模型,分别实现了独立校准方式采用最小二乘法求解误差模型参数的算法,和非独立校准方式基于知识(Knowledge-based)与卡尔曼滤波器的在线训练模型参数求解算法,通过大量实验讨论了算法的适用条件,验证了该误差模型的合理性和参数求解算法的有效性。3.行人无缝定位算法实现为了实现行人室内外无缝定位技术,本文设计了一种三模式的定位机制:GPS定位模式--当GPS信号质量好时直接利用GPS定位结果;PDR定位模式--无GPS信号时完全采用PDR定位结果;GPS和PDR混合定位模式--信号质量不好时GPS与PDR定位结果通过融合Kalman滤波器得出最终定位信息。分别实现了在无GPS信号时基于加速度和航向与基于EMG和航向的两种PDR算法,通过实验验证了在环境磁干扰较少的情况下PDR算法典型定位精度能够达到甚至优于目前大多数系统的水平。基于一组在城市峡谷中室内外定位实验的数据,重点研究了GPS和PDR混合定位模式的算法,包括GPS定位质量评估和融合滤波器的设计,并探讨了与其它技术结合对于进一步提高定位精度的潜力。更多还原

【Abstract】 As navigation function becomes a more and more common application in mobile phones and LBS (Location-Based Service) grows rapidly, requirements are increasingly urgent on the accuracy, reliability and continuity of pedestrian navigation and positioning technology. However, as a primary positioning means in pedestrian navigation, the U.S. GPS (Global Positioning System) receiver can’t provide accurate results and even fails in positioning under complex urban canyons and indoor environments due to signal attenuation, interference or blockage, etc, in which most of personal users’activities exactly happen. In order to obtain continuous navigation, GPS must be augmented by other positioning technologies, such as self-contained sensors, WLAN (Wireless Local Area Network), mobile cellular networks, RFID (Radio Frequency IDentification), Pseudolites, etc. Integrating GPS with self-contained DR sensors is autonomous and doesn’t need extra infrastructure, or a fingerprint database, so that it attracts extensive attentions and intensive investigation.Based on a self-developed low-cost Multi-Sensor Positioning platform (MSP), which includes a GPS receiver and two self-contained sensors (a 2-axis digital compass and a 3-axis accelerometer), this dissertation investigates if the GPS receiver can’t provide accurate and continuous positioning information, how to obtain a pedestrian’s speed and heading from self-contained sensors and calculate his/her position through combined Kalman filters, for achieving a seamless outdoor/indoor pedestrian positioning solution.After deeply studying on various seamless pedestrian positioning algorithms, taking into account that the performance of low-cost sensors in MSP can’t meet the accuracy requirement of traditional inertial navigation mechanism, the dissertation selected Pedestrian Dead Reckoning (PDR) algorithm to assist the positioning in the complex environment that GPS performance is poor, and focused on the following three aspects:1. Step detection and step length estimation algorithms: based on physiological characteristics of a pedestrian gait, referring to traditional step detection and step length estimation methods, a step detection algorithm is realized using accelerometer’s signals, combing three methods: sliding-window, zero-crossing and peak detection, and a 1-parameter step length estimation model is chosen. In addition, a novel biomedical signal, Electromyography (EMG) is firstly introduced into personal navigation, for detecting the pedestrian’s step occurrences and corresponding step lengths, and the details of the proposed method are presented, including setup of EMG electrodes, signal pre-processing procedure, algorithms of stride detection and stride length estimation. Several experiments were carried out to validate step detection methods and evaluate the precision of step length estimation models separately.2. Digital compass heading calibration algorithm: it’s always a key issue in PDR algorithms to obtain accurate heading, as well as a long-term problem in pedestrian positioning research. Since a 2-axis digital compass can’t compensate tilt error by itself, and is vulnerable to various errors, such as hard and soft iron effects, biases, scale factors and oscillation of the pedestrian’s body when walking, a simulation method is utilized to study the impact of each error on outputs of digital compasses. Fully considering the easy-to-use requirement on heading calibration methods, a unified heading error model is proposed, which includes all the possible and predictable errors. Two different algorithms are used to solve the error model’s parameters. One of these algorithms requires an independent calibration procedure using least square method, and the other one is non-independent using Knowledge-based and Kalman filter for online training the model’s parameters. The application condition and feasibility of the calibration approach has been discussed and validated through extensive field tests.3. Seamless pedestrian positioning algorithm: to achieve a seamless outdoor/ indoor pedestrian positioning solution, this dissertation proposes a three-mode positioning mechanism: GPS mode- when GPS signal is good, final positioning results derive from GPS receiver directly; PDR mode- when there is no GPS signal, final positioning information are obtained from PDR results; GPS/PDR hybrid mode- when GPS signal can be received but not good, the final results are the ones combined by both GPS and PDR positioning outputs through Kalman filter. When GPS signal is not available, the acceleration-based and EMG-based PDR algorithms are realized separately with the aid of the heading from digital compass, and verified in several field tests conducted in the environments including less magnetic disturbances. The results demonstrate that the typical positioning performance of the MSP can meet and even exceed the common level in most of existing pedestrian positioning systems. Based on a set of data collected in a field test conducted in the urban canyon, the GPS/PDR hybrid positioning algorithm has been developed, including quality evaluation of GPS positioning performance and design of fusion filter, and the potential of combining the algorithm with other techniques is explored in order to further improve the positioning accuracy.更多还原