【泡泡一分钟】AC / DCC：Visual SLAM的动态摄像机群集的准确校准

每天一分钟，带你读遍机器人顶级会议文章

标题：AC/DCC : Accurate Calibration of Dynamic Camera Clusters for Visual SLAM

作者：Jason Rebello Angus Fung and Steven L. Waslander

来源：2020 IEEE International Conference on Robotics and Automation (ICRA)

编译：张宁

审核：柴毅，王靖淇

这是泡泡一分钟推送的第 642 篇文章，欢迎个人转发朋友圈；其他机构或自媒体如需转载，后台留言申请授权

摘要

为了在动态相机集群（DCC）中的各个相机之间关联信息，需要确定一个准确的随时间变化的外部校准转换集。先前的校准方法仅依赖于从已知基准目标收集测量值，这限制了校准精度，因为实现了万向节的激励不足。在本文中，我们通过收集云台整个配置空间的测量值来提高DCC校准精度，并将像素重投影误差减小10倍。我们使用姿势环误差优化方法对任意数量的摄像机和未知关节角度之间的校准参数执行联合优化，从而避免了重叠的视场。我们在仿真中测试了我们的方法，并针对不同级别的摄像机固有噪声和关节角度噪声提供了校准灵敏度分析。此外，当关节角度值未知时，我们还提供了对标定中简并参数的新颖分析，从而避免了无法唯一恢复标定的情况。校准代码将在https://github.com/TRAILab/AC-DCC上提供。

图1.一个动态摄像机组，它由两个静态摄像机和一个动态摄像机组成，该摄像机连接到210 RTK DJI Matrice上的3-DOF云台。

图2.用于配置测试的多维数据集目标环境

图3.灵敏度分析测试，比较了不同像素水平和关节角度噪声的重投影误差。

Abstract

In order to relate information across cameras in aDynamic Camera Cluster (DCC), an accurate time-varying setof extrinsic calibration transformations need to be determined.Previous calibration approaches rely solely on collecting measurements from a known fiducial target which limits calibrationaccuracy as insufficient excitation of the gimbal is achieved. Inthis paper, we improve DCC calibration accuracy by collectingmeasurements over the entire configuration space of the gimbaland achieve a 10X improvement in pixel re-projection error. Weperform a joint optimization over the calibration parametersbetween any number of cameras and unknown joint anglesusing a pose-loop error optimization approach, thereby avoidingthe need for overlapping fields-of-view. We test our methodin simulation and provide a calibration sensitivity analysisfor different levels of camera intrinsic and joint angle noise.In addition, we provide a novel analysis of the degenerateparameters in the calibration when joint angle values areunknown, which avoids situations in which the calibrationcannot be uniquely recovered. The calibration code will be madeavailable at https://github.com/TRAILab/AC-DCC.