Reference markers in source and target footage for automated or manual tracking — compositors solve camera motion precisely. Minimum three per frame.
You set control points to track the camera — and with that, you lay the foundation for any match move, any camera solve in post. Essentially, you mark corresponding points in the source footage and the target footage: identical locations in space and time that the tracking algorithm then uses as anchor points to calculate the exact movement and geometry. This works automatically with good features (high contrast, sharp corners), but as soon as it gets complex — missing markers, motion blur, lighting changes — you have to manually correct or add more points.
The minimum number is three points per frame. Why? With three non-collinear points, a 2D homography can be calculated. If you need a more robust solution or 3D camera data, you set more — five to ten are common. The more points you place, the more stable the solution becomes, but also the more tedious the correction. The trick is to choose features that are distributed across the entire frame and remain visible for multiple frames. A corner on a building is better than a spot on a T-shirt that deforms.
In practice, you work with two approaches in parallel. With automatic tracking, you start with good points (usually four corners of a square), and the tracker follows them frame by frame. This is fast but prone to issues: occlusion, motion blur, or focus plane shifts interrupt the tracking. Then you have to manually correct — you click in each problematic frame, reposition the point. This is the unglamorous, time-consuming craft. Experienced compositors quickly see where things go wrong: the points drift too fast or stop while the camera continues to move.
A common mistake is placing points on moving objects or shadow edges. You need static geometry or at least corresponding points that move predictably (like a car you've rigged beforehand). Also, pay attention to calibration: if you set control points in HD for a 4K shoot, the resolution won't match later. Modern tools like Mocha or After Effects offer planar tracking — this significantly reduces the point overhead because a flat surface is sufficient. But for 3D tracking for camera solve, you still need actual spatial points.