知识卡片

普通读书笔记卡 · 9054.b · 生成模型:GPT-6

SfM初始化需要可靠视差

理解等级:中级

可从一对有足够重叠与基线的图像开始,以RANSAC剔除错误对应,估计相对几何并三角化初始点。单纯选最多匹配的图对不保证好初始化:纯旋转或极小基线仍难恢复深度。初始几何与后续束调整共同决定重建能否落入合理解。

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

      There are many different ways in which one can attempt to minimize the loss from equation (44.5). One standard approach is to do it incrementally starting with only a pair of images. We use the matches between those two images to compute the fundamental and essential matrices, while accounting for outliers using RANSAC. We can then compute an initial 3D reconstruction of the matched points

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第44章《多视图几何与运动恢复结构》(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/c044.xhtml) - 结论依据:摘录中关于SfM初始化需要可靠视差的定义、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:There are many different ways in which one can attempt to minimize the loss from equation (44.5). One standard approach is to do it incrementally starting with only a pair of images. We use the matches between those two images to compute the fundamental and essential matrices, while accounting for outliers using RANSAC. We can then compute an initial 3D reconstruction of the matched points