← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 9035.e · 生成模型:GPT-6

单应重投影优化需先去齐次

理解等级:中级

DLT代数残差不直接表示像素误差。可用它初始化,再最小化p2与去齐次化(Hp1)之间的几何残差;需要时考虑双向误差。每个候选都要在正确坐标和尺度下比较,不能直接把齐次三向量的欧氏差称作像素重投影误差。

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

      In practice, to get accurate results, it is useful to also minimize the reprojection error after you initialize with the DLT solution since ∥Ah∥2 is meaningless geometrically. 41.3.2 Robust Model Fitting: Random Sampling with Consensus To perform the image stitching in the section above, we had to find the homography parameters that best described the transformation from one image to another. Because the detection of

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第41章《单应性》(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/c041.xhtml) - 结论依据:摘录中关于单应重投影优化需先去齐次的定义、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:In practice, to get accurate results, it is useful to also minimize the reprojection error after you initialize with the DLT solution since ∥Ah∥2 is meaningless geometrically. 41.3.2 Robust Model Fitting: Random Sampling with Consensus To perform the image stitching in the section above, we had to find the homography parameters that best described the transformation from one image to another. Because the detection of