【电信学】【2005.01】用于视觉辅助惯性导航的随机约束

本文为美国MIT（作者：David D. Diel）的硕士论文，共110页。

本文提出了一种基于特征约束的改进的惯性导航方法。提出的方法延长了人类或车辆在缺乏GPS的环境中正确导航的时间。我们的方法与现有的导航系统集成得很好，因为我们调用了表示各种可用硬件的通用传感器模型。惯性模型包括偏差误差、比例误差和随机游动误差。可以使用任何相机和跟踪算法，只要视觉输出可以表示为从传感器主体上的已知位置延伸的射线向量。采用改进的线性卡尔曼滤波器进行数据融合。与传统的同步定位和绘图（SLAM/CML）不同，我们的状态向量只包含与位置相关的惯性传感器误差。这种选择允许不确定性由协方差矩阵正确表示。我们不使用特征坐标来增强状态。取而代之的是，图像数据在时间和空间上对宽基线产生随机极线约束，从而提高了IMU误差状态的可观测性。约束导致相对残差和相关的相对协方差，部分由状态历史定义。导航结果是使用高质量的合成数据和真实的鱼眼图像展示的。

This thesis describes a new method toimprove inertial navigation using feature-based constraints from one or morevideo cameras. The proposed method lengthens the period of time during which ahuman or vehicle can navigate in GPS-deprived environments. Our approachintegrates well with existing navigation systems, because we invoke generalsensor models that represent a wide range of available hardware. The inertialmodel includes errors in bias, scale, and random walk. Any camera and trackingalgorithm may be used, as long as the visual output can be expressed as rayvectors extending from known locations on the sensor body. A modified linearKalman filter performs the data fusion. Unlike traditional SimultaneousLocalization and Mapping (SLAM/CML), our state vector contains only inertialsensor errors related to position. This choice allows uncertainty to beproperly represented by a covariance matrix. We do not augment the state withfeature coordinates. Instead, image data contributes stochastic epipolarconstraints over a broad baseline in time and space, resulting in improvedobservability of the IMU error states. The constraints lead to a relativeresidual and associated relative covariance, defined partly by the statehistory. Navigation results are presented using high-quality synthetic data andreal fisheye imagery.

1. 引言
2. 研究方法
3. 惯性测量
4. 视觉测量
5. 极线约束滤波器
6. 实验
7. 分析
8. 结论
   附录A 系统参数权衡
   附录B 映射参考
   附录C 旋转参考

下载链接：

https://089u.com/f/1850492-503202031-143b28

（访问密码：3660）

更多精彩文章请关注公众号：

标签：误差,惯性导航,2005.01,约束,协方差,随机,传感器,附录,data
来源： https://blog.csdn.net/weixin_42825609/article/details/119045924