Frontier papers

COLMAP:把照片变成可测量的稀疏重建

Structure-from-Motion Revisited

Schönberger, Frahm · 2016 · arXiv:1604.03755

Foundation paper · permanent
2016
The question

无人机拍回来一堆无序照片,能否自动恢复相机位姿与稀疏点云?

What they did

增量式 SfM:特征匹配 → 几何验证 → 增量注册 → 光束法平差。

Evidence

在多个大规模数据集上鲁棒性与完整性领先,成为摄影测量的事实基线。

Do not over-trust

重叠不足、弱纹理、运动模糊会让注册失败;位姿错一帧,后面整块模型都歪。

Limits

需要足够重叠与纹理;尺度需外部基准;反光与薄结构匹配差。

In-class task

用一组航拍图跑 COLMAP,统计注册成功率,并用已知基线长度检查尺度。

A foundation paper is the landmark work for a method and stays permanently: it explains why the method works, and is never replaced for being old.
Paper Related chapters: 点云索引:百万点怎么查 Back to learning