Skip to content

Tracking & Triangulation

This section explains the two core algorithms behind LUC3D's 3D reconstruction: Tracking (linking the same individual across frames and views) and Triangulation (fusing 2D labels into 3D points).

Where the User Guide covers how to use these features, this section covers how they work under the hood.

Tracking

Tracking is the process of establishing correspondence — deciding which 2D instances across different frames and camera views belong to the same physical subject. These correspondences (instance groups) are what make triangulation possible: a 3D point can only be reconstructed once LUC3D knows which 2D labels describe the same keypoint.

→ Read the Tracking page

Triangulation

Triangulation fuses the grouped 2D labels into a single 3D estimate for every keypoint, then projects that estimate back into each camera so you can verify the reconstruction against the original labels (the reprojection).

LUC3D offers two methods:

  • DLT (Direct Linear Transform) — fast, linear, single-shot.
  • Refined (Ref) — slower and iterative; helps when one view's 2D point is clearly wrong, otherwise about the same as DLT. Its reported error is never larger than DLT's, but that error is measured on the points it fits.

→ Read the Triangulation page

Temporal Smoothing

Temporal Smoothing is an optional last step: it filters each node's 3D over time to remove one-frame spikes and jitter.

→ Read the Temporal Smoothing page