Camera tracking for static shots

The film is made entirely of locked-off shots: dialogue, reactions and ambient moments with the camera on a tripod. Static does not mean simple — the robot and the CG city still needed exact perspective and scale.

Lens distortion in 3DEqualizer

We tried 3DEqualizer for full solves on the first hero shots, then kept it for one job only: lens distortion, where it gives the best maps.

1. Each lens calibrated from grid charts and lens data shot on set.
2. Radial and tangential distortion analysed and exported as maps for every focal length.
3. Undistort / redistort pairs, so every CG render returns to plate space without loss.
4. Maps checked in Nuke against tracking markers and plate geometry — including cameras solved elsewhere.

DistortedUndistortedDrag to compare

fSpy for 80% of the shots

With a tight schedule, about 80% of the cameras were solved in fSpy, an open-source tool that reconstructs a camera from a single frame. It works best on locked-off shots with strong architectural lines and little depth variation.

1. Key frame extracted from the final plate.
2. Vanishing points and horizon set with perspective guides.
3. Scale defined from a known size — a door, a table.
4. Camera exported to Blender or Maya and checked with proxy geometry in Nuke.
5. Tilt and focal length fine-tuned by hand so the shadows land right.

Setup time per shot dropped from several hours to under 15 minutes, with the precision the robot needed.

Validation

Every camera, from either tool, was tested in Nuke by projecting geometry onto the plate: grid projections for distortion, robot proxies for ground contact, test renders for reflections and shadows. Anything that drifted was re-solved or adjusted in Maya.