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
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.
© SailiTech · Printed at: