基于惯性传感器的行人航位推算算法的研究与实现

发布时间：2018-03-25 20:29

本文选题：行人航位推算　切入点：室内地图匹配　出处：《电子科技大学》2017年硕士论文

【摘要】：行人航位推算常用于估算用户在室内环境的位置,然而,传感器测量数据偏差,传统航位推算算法中误差的积累,对最终的定位结果影响很大;其次,用户持握方式是多变的,很难建立起传感器数据与持握方式之间的准确关系,难以根据数据特征进行分类;再有,行人行走轨迹可能会出现偏移较大、穿墙而过的错误现象。针对上述问题,本文提出一个改进的行人航位推算算法,在穿零检测法基础之上,提出上升穿零与下降穿零两种情况下的步数统计方法;接着,综合五种变量:性别、身高、步频、波峰加速度、波谷加速度,提出新的步长计算方法,通过多元变量线性回归分别训练男、女数据样本得到该模型,接着使用粒子滤波,修正间歇性跳跃的步长结果;然后,根据磁力计方向角和陀螺仪相对转向角度的线性相关性,线性拟合得到基于二者角度变化值的方向估算模型,以弥补传统航位推算算法中单独使用磁力仪或陀螺仪进行方向估算产生的大幅偏差;此外,针对不同的持握方式,根据其加速度信号,利用小波变换提取运动特征,奇异值分解进行特征降维,通过支持向量机进行分类训练,达到准确判断不同持握方式的目标,并根据特定持握方式下,用户对导航轨迹不敏感的特点,统计步数与步长;最后,本文提出室内地图建模与匹配方法,先对室内空间进行抽象、建模,再通过兴趣点匹配、穿墙检测、方向修正等步骤限制用户运动轨迹,使其与真实轨迹更加贴近。实验结果表明,计步阶段,穿零检测法平均误差0.8%,传统波峰检测法平均误差11.6%;本文提出的步长计算误差3.5%,相比于已有的身高-步频模型误差8.63%,以及波峰-波谷根式模型10.84%,有显著提升;方向估算结果90%以内的样本误差都在20°以内,对比传统的磁力计、陀螺仪有明显的优势;使用SVM进行持握方式判断的准确率高达95.62%,而贝叶斯分类的准确率只有82.31%。最后地图匹配的实验中,绘制轨迹克服了漂移、穿墙等问题,改进的航位推算算法平均误差1.48m,绘制的轨迹能反映行走路线。本文出的改进的航位推算算法及地图匹配在室内定位中大大提高了定位精度,具有较高实用性。
[Abstract]:Pedestrian dead-reckoning is often used to estimate the user's position in the indoor environment. However, the error accumulation in the traditional dead-reckoning algorithm has a great influence on the final positioning results. It is difficult to establish an accurate relationship between the sensor data and the holding mode, and it is difficult to classify according to the characteristics of the data. In addition, the pedestrian track may have errors of deviation and passing through the wall. In this paper, an improved algorithm is proposed to calculate the number of steps on the basis of the zero-piercing method, and then the statistical method of step number is proposed under the condition of zero passing through rise and zero down, and then five variables are synthesised: gender, height, step frequency, and so on. Wave peak acceleration, trough acceleration, a new calculation method of step size is proposed. The model is obtained by training the male and female data samples separately by linear regression of multivariate variables, and then using particle filter to modify the step result of intermittent jump; then, According to the linear correlation between the direction angle of the magnetometer and the relative steering angle of the gyroscope, a direction estimation model based on the change value of the two angles is obtained by linear fitting. In order to make up for the large deviation caused by direction estimation by using magnetic force instrument or gyroscope alone in the traditional dead-reckoning algorithm, in addition, according to the acceleration signal of different holding mode, the motion feature is extracted by wavelet transform. The singular value decomposition (SVD) is used to reduce the dimension of the feature, and the SVM is used to classify and train to judge the target of different holding mode accurately. According to the characteristic that the user is insensitive to the navigation track, the step number and step size are counted. Finally, this paper proposes a method of indoor map modeling and matching. Firstly, the indoor space is abstracted, modeled, and then the user's trajectory is restricted by the matching of interest points, the detection of the wall, the direction correction and so on. The experimental results show that, in the step stage, The average error of zero-penetrating detection method is 0.8, the average error of traditional wave peak detection method is 11.60.The error of step size calculated in this paper is 3.50.Compared with the existing height and step frequency model error 8.63, and the crest and trough root model 10.84, it has a significant improvement. The sample error within 90% of the estimated direction is less than 20 掳. Compared with the traditional magnetometer, gyroscope has obvious advantages. The accuracy of using SVM to judge the holding mode is 95.62, but the accuracy of Bayesian classification is only 82.31. Finally, in the experiment of map matching, drawing track overcomes the problems of drift and wall, etc. The average error of the improved algorithm is 1.48 m and the plotted track can reflect the walking route. The improved algorithm and map matching in this paper greatly improve the positioning accuracy and have a higher practicability in indoor positioning.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP212.9

【参考文献】