【24h】

Integration of Vision and Navigation

机译:视觉与导航的整合

获取原文

摘要

Navigation using only GPS and IMU sensors isubiquitous for vehicle motion applications and the theoryand mechanizations are well documented. Numerouspapers further describe the use of imagery from anonboard camera to prevent drift of navigation errors overtime. Other papers describe using GPS and IMU foraiding preparation of georegistered imagery mosaicswhere vehicle navigation is secondary. Finally, there arevarious procedures that use only planar imagery, withoutassistance from motion sensors, to recreate 3Dgeoregistered scenes that can be viewed from anygeometry. All of these methods mix and match motionsensor data and imaging data to prepare some means ofgenerating a user spatial awareness through thecomputation of position and attitude (pose) at the discretecamera events.This paper presents a unified theory for merging thesespatial sensing modalities based upon development of aMarkov process representation of the pose evolution. Thisapproach is an extension of the IMU and GPS navigationsolutions that also make use of a Markov process fortransitioning between external observations. This paperdescribes how overlapped image sequences can be castinto a similar stochastic Markov representation thatmerges all 3D information from camera sensors with datafrom motion sensors. A sequence of overlapped imageframes forms a Markov process for evolving pose that canbe merged with traditional IMU and GPS stochasticmodels. Out-of-sequence overlapped images act asobservations to the underlying image-derived Markovprocess so that the complete overlapped image set isintegrated in an optimal manner. Kalman filter/smootherprocedures can be applied so that the navigation andscene geospatial content are generated conditioned uponall IMU, GPS, and image measurements. The statisticalaccuracy of all estimated geometry information isnaturally provided from the estimation results.This paper describes this integrated approach for vehicleplatforms that include a digital camera, GPS, and aMEMS IMU that is rigidly affixed to the camerastructure. All processing occurs within a single recursivealgorithm with selectable parameters to weight thevarious measurements. Covariance analysis indicatingestimation accuracy is used to define relative benefits ofthe contributing sensors and for specification of imagecollection strategies.
机译:仅使用GPS和IMU传感器进行导航 在车辆运动应用和理论中无处不在 机械化有据可查。很多的 论文进一步描述了图像的使用 车载摄像头,以防止导航误差漂移 时间。其他论文介绍了将GPS和IMU用于 协助准备地理配准的图像镶嵌 车辆导航为次要位置。最后,有 仅使用平面图像的各种程序,而没有 运动传感器的协助,以重新创建3D 可以从任何位置查看的地理注册场景 几何学。所有这些方法混合并匹配运动 传感器数据和成像数据准备一些手段 通过 离散位置和姿态(姿势)的计算 摄影机事件。 本文提出了一个统一的理论来合并这些 空间感知模态的发展 姿势演化的马尔可夫过程表示。这 方法是IMU和GPS导航的扩展 还利用马尔可夫过程的解决方案 外部观察之间的过渡。这篇报告 描述了如何投射重叠的图像序列 变成类似的随机马尔可夫表示, 将来自相机传感器的所有3D信息与数据合并 来自运动传感器。一系列重叠的图像 框架形成了可以演化的姿势的马尔可夫过程,可以 与传统的IMU和GPS随机合并 楷模。乱序重叠的图像充当 对基础图像衍生的马尔可夫的观测 处理,以便完整的重叠图像集为 以最佳方式集成。卡尔曼滤波器/平滑器 可以应用程序,以便导航和 场景地理空间内容是根据以下条件生成的 所有IMU,GPS和图像测量。统计 所有估计的几何信息的准确性为 根据估算结果自然提供。 本文介绍了这种针对车辆的集成方法 平台,其中包括数码相机,GPS和 固定在相机上的MEMS IMU 结构体。所有处理都在单个递归中进行 带有可选参数的算法 各种测量。协方差分析表明 估计精度用于定义以下方面的相对利益: 贡献传感器和图像规格 收集策略。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱:kefu@zhangqiaokeyan.com

京公网安备:11010802029741号 ICP备案号:京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有
  • 客服微信

  • 服务号