Camera tracking, also called matchmoving, works by analysing parallax: the way objects at different distances appear to shift by different amounts as the camera moves through space. Nearby objects move a lot in the frame while distant objects barely move at all. This difference is what allows the solver to calculate depth and reconstruct the 3D scene.
Think of it like your own eyes. They are spaced a few centimetres apart, and each eye sees the world from a slightly different position. Your brain compares those two views and uses the difference to perceive depth. Camera tracking works the same way, except instead of two eyes it uses many frames captured from slightly different positions over time.
A tripod shot with only pan or tilt does not produce parallax. When the camera rotates in place, every object in the scene, near and far, shifts by the same amount in the frame. There is no depth information for the solver to work with, and the reconstruction will fail or produce unusable results.
What works:
- Walking or sliding the camera sideways, forward, or backward through the scene
- Arcing or orbiting around a subject
- Handheld movement where the camera physically travels through space
- Dolly, crane, or gimbal shots with actual translation
What does not work:
- Tripod pan and tilt: the camera stays in one spot and only rotates
- Zooming in and out without moving the camera body
- Nodal pan (rotating around the lens entrance pupil)
Even a small, steady physical movement is enough to give the solver the parallax it needs. The key rule is simple: move the camera's position, not just its angle.
The exceptions are the two shot-specific trackers added in 1.0.7 (Max): the Tripod Tracker is built precisely for rotation-only pans, tilts, and rolls, and the Planar Tracker follows a flat surface such as a screen or sign regardless of camera motion. Neither needs parallax, because neither reconstructs 3D depth from it.