基于低成本惯性传感器的井下人员定位关键算法研究

Currently, inertial sensor based indoor navigation is a research focus and studying it is a trend. As to mine laneway, its roads are rugged and filled with detritus. Persons under the mine are in different movement states such as walking, running, sidling, ascending or descending the slop, taking the lift or electric locomotive, and creeping in case of mine accidents. There is strong magnetic interference under the mine because of too many cables, metal pipelines and pieces of facilities there. The map of mine laneway is required to rapidly update as the mine is exploited in a large scale. In the frame of combining integral operation based on constant/zero speed check and pedestrian dead reckoning, the data on persons’ movement in a noisy environment is used, and blind source separation to the extraction of spine wave for gait analysis is applied to explore the calculation method of step length based on response analysis system and persons’ motion state. The map of the mine is fast constructed by using such technologies as laser scanning and map stitching. In the case of the straight or winded road, adaptive Kalman filtering model based on fading factor is built with the technologies such as orientation constraint, magnetic interference detection and compensation, and blue tooth Beacon absolute localization. Based on this, low-cost inertial sensor based accurate localization system of persons under the mine is completed so as to provide support for the exploitation of the mine and decision-making.

基于惯性传感器的室内导航是当前的研究热点和趋势，但对于矿井巷道，其路面往往坑洼不平，甚至布满碎石；井下人员的运动状态更为复杂，走路、跑步、侧行、上下坡、乘坐电梯/电机车、发生矿难时还有可能爬行等；井下各种设施、线缆、金属管道繁多，有很强的磁干扰；随着矿井向深部高强度、大面积的开采，巷道地图也需要快速更新。鉴于此，拟从高噪声的井下人员步态数据入手，利用盲源分离技术抽取出稳定步态特征的脊柱波做步态分析，根据人员运动状态，利用反应分析体系探寻自适应的步长计算方法；通过激光扫描、地图拼接等技术实现井下地图的快速构建；在基于常速/零速检测的积分运算及基于航迹推算两种定位机制的融合框架下，通过直/曲巷道地图方向约束、磁干扰的检测及补偿、蓝牙Beacon绝对定位等技术实现自适应卡尔曼滤波模型，探索基于低成本惯性传感器的井下人员长时间、高精度的定位系统实现方法，为矿井安全开采、辅助决策等方面提供技术保障。

本项目实现了复杂环境下高噪声步态数据检测、运动模型分析、步长参数确定以及矿井巷道地图快速更新等技术；建立基于盲源分析的脊柱波提取与步态检测理论；基于反应分析体系的稳定的步长计算方法；基于十二位置的传感器误差标定及磁力计噪声干扰检测与排除算法；基于二维激光扫描、地图拼接等技术实现井下地图的快速构建算法；基于直/曲巷道地图方向约束算法；基于步态分析基础上的零速检测算法；基于积分计算和航迹推算等不同技术方法的融合定位模型；基于UWB/IMU融合定位模型等。所取得的主要成果有：足部IMU数据经振动算子修正以后，距离计算精度有了较大提高，对于200 m的采样距离，Weinberg步长模型的部分姿态距离累积误差仅为1~2 m。在融入地磁信息后，利用回环检测结果，通过图优化算法进行优化后的轨迹的累计误差被显著地抑制，表现出良好的精度，受到回环约束的路径平均误差控制在2.15m以下。在PDR与UWB的EKF融合算法中，通过Mahalanobis 距离去抑制UWB数据异常对定位结果的影响。实验结果表明，对于UWB数据的间歇噪声、连续噪音、连续中断等异常，该算法均有较强的鲁棒性。在UWB与IMU融合在复杂运动条件下的计算，实现一种基于运动约束的迭代鲁棒扩展卡尔曼滤波定位（Constraint Robust Iterate Extended Kalman Filter,CRIEKF）算法。该算法能克服扩展卡尔曼滤波（Extended Kalman Filter,EKF）对非高斯噪声的天生缺陷，及鲁棒扩展卡尔曼滤波（Robust Extended Kalman Filter,REKF）算法仅利用先验信息处理非高斯噪声的缺点，在每一步的迭代中基于系统的后验估计更新观测值的协方差，然后根据得到的协方差更新系统的后验分布，显著降低了非高斯噪声对定位精度的影响。同时，在计算中引入了运动约束，如零速约束、伪速约束、平面约束等，进一步平滑了定位结果。初步实验结果表明，基于CIREKF算法的UWB于IMU融合的小车定位，在NLOS环境下，平均定位精度约0.21米。