知识卡片

普通读书笔记卡 · 9026.d · 生成模型:GPT-6

低残差不独证参数可信

理解等级:中级

书中办公室以12个手工对应得到约12.3像素平均重投影误差,并恢复约171.8cm相机高度;这些是该历史示例的结果。像素误差相对图像较小,也不证明各内外参精确或能迁移到其他画幅。需检查测量误差、视野覆盖、外推区域与独立参照。

这张卡片看不懂?

已有反馈 · 0 条

    还没有人提交反馈。

    在整书笔记中的位置

    原书出处 · 1 处

    • · 查看原书依据

      The matrix has been scaled so that the bottom-right element is a 1. Using this matrix, the mean reprojection error is 12.3 pixels, which is reasonably small for an image of size 4,032 × 3,024 pixels. That is, the reprojection error is < 0.5 percent of the image height.

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第39章《相机建模与标定》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)/_epub-src/OPS/xhtml/c039.xhtml) - 结论依据:摘录中关于低残差不独证参数可信的定义、比较或条件支撑卡片;原书表述不严谨的范围已在覆盖记录逐项限定,不把案例结果外推为普遍结论。 - 原始内容:The matrix has been scaled so that the bottom-right element is a 1. Using this matrix, the mean reprojection error is 12.3 pixels, which is reasonably small for an image of size 4,032 × 3,024 pixels. That is, the reprojection error is < 0.5 percent of the image height.