From photos and video to measurable 3D geometry.
I build 3D reconstruction pipelines in Python: camera calibration, pose estimation, triangulation, point clouds and meshes — with a clear technical report on the accuracy you can expect.
What it can do
Built on classical multi-view geometry, validated on your own data.
Camera calibration
Estimate intrinsics and lens distortion from chessboard images or from your own data, with the reprojection error reported.
Two- and multi-view geometry
Fundamental/essential matrix, camera pose recovery, triangulation and bundle-adjustment refinement.
Point clouds & meshes
Filtering, alignment, meshing and export to PLY/OBJ/CSV for use in your own tools.
Measurement
Scripts that measure distances, volumes or dimensions from the model, with an uncertainty estimate.
Honest reporting
I state what can and cannot be measured, and how accuracy depends on baseline, resolution and image quality.
Reproducible code
Clean Python with run instructions so you can re-run it on new data. No lock-in.
See how I think
Two technical articles, with equations, figures and runnable code, explaining the fundamentals behind this service.
Camera calibration from a chessboard: Zhang's method
How a planar pattern yields the intrinsic matrix in closed form, why at least three views are needed, and a Monte-Carlo experiment on noise and number of images.
3D reconstruction · 2026-11-04The eight-point algorithm and why normalisation matters
Estimating the fundamental matrix with an SVD, why raw pixel coordinates make the problem ill-conditioned, and a numerical experiment on Hartley normalisation.
3D reconstruction · 2026-11-04Triangulation: from two views to a 3D point
Why two noisy rays never intersect, how the Direct Linear Transform solves it with an SVD, and a numerical experiment on how camera baseline controls accuracy.
What I need from you
- Example images or video and the object/scene scale
- Required accuracy and output format
- Camera information if available (model, focal length, chessboard images)
Not a fit for: requirements that need certified survey-grade accuracy, or data with too few views or too much blur for a reliable reconstruction. I will tell you so plainly after seeing sample data.
Send me a few sample photos
Describe the goal in two or three sentences; I will reply with clarifying questions and a fixed-scope proposal.