Camera Tracker

Camera Tracker is for recovering the motion of the physical camera rather than following one point or one planar surface. It measures many features across the frame, builds overlapping tracks through time, and solves for the virtual camera path and orientation that best explain the observed 2D motion.

The node has two physical inputs. Orange Background receives the plate used for tracking. White Occlusion Mask excludes regions that should not participate in the analysis, such as a moving person, vehicle, changing screen, reflection or other foreground motion. White areas of the mask are not tracked.

The working loop does not end at Auto Track. Reliable camera solves are usually iterative: Track, Solve, filter, Solve again, manually clean, and solve again. Long overlapping tracks are more useful than a huge collection of short random points, and false tracks can strongly bias the reconstructed camera.

The Camera tab defines the physical shooting model: Focal Length, Film Gate, aperture, gate fit, optical center and pixel aspect. The more real camera information you know, the fewer variables the solver must estimate. If focal length is unknown it can be refined; if it is known, entering it explicitly is usually safer.

The Solve tab is both a solver and a diagnostic workspace. Average Solve Error is the pixel distance between 2D tracks and reconstructed 3D locators reprojected through the solved camera. Blackmagic notes that below about one pixel is generally useful for Full HD and roughly 0.6–0.8 pixels is excellent, but the number must still be judged together with local reprojection and the target resolution.

Lens Parameters should only be enabled when real lens distortion is present and has not already been removed upstream. Camera Tracker can model radial and tangential distortion, but every extra lens variable makes the solve less constrained. For strongly distorted wide-angle plates, a clean undistort-solve-redistort workflow can be easier to reason about.

Export is not a second socket on Camera Tracker. The Export button creates separate 3D nodes: an animated Camera3D, reconstructed PointCloud3D, Shape3D ground plane, Merge3D and Renderer3D, with optional lens-distortion and image-plane elements. Update Previous Export refreshes those existing nodes after a new solve without rebuilding the downstream graph.

3D Scene Transform does not improve or modify the camera solve. It only reorients the exported scene so Origin, Orientation and Scale are convenient for downstream work. Solve first, align second, then validate with a simple Text3D or cube. If the object swims against the plate, return to track cleanup rather than hand-animating the 3D object to hide the error.