Crime Scene Reconstruction

Sparse SfM

Real structure-from-motion from multiple photos — sparse point cloud + camera poses, not a forensic tool

Upload 2-6 photos of the same static scene, taken while walking around it

Optional: calibrate to approximate real-world units

Click two points in photo 1 whose real-world distance you know (e.g. two corners of a table), then enter that distance. Without this, the reconstruction is in arbitrary relative units — with it, still only approximate (no camera calibration is performed).

What this doesn't do: this is sparse structure-from-motion, not a dense 3D model — it reconstructs matched keypoints only. There is no bundle adjustment or loop closure, so pose accuracy degrades with more photos, and no camera calibration is performed, so shape and scale are approximate even with the optional distance calibration. This is an educational demonstration of the real technique behind photogrammetry/SfM tools like COLMAP, not a forensic-grade measurement tool.