A 3D track that fails in review usually fails in one of three specific ways. The track wanders off the plate over time. The track loses sync with the plate by a few frames. The track holds the foreground but loses depth in the background. Each of these is a different problem with a different root cause, and the working matchmove artist diagnoses by symptom — what’s the visible failure — before reaching for the fix.
The artist who’s only learned to “improve the solve” by adding more tracking points or tweaking the iteration count fixes some tracks and breaks others, because the same intervention has different effects on different failure modes. The artist who diagnoses first works faster and produces tracks that hold under stabilization tests.
This post is the working matchmover’s guide to the three primary failure modes. What they look like, why they happen, and what the right intervention is for each.
Failure Mode 1: Drift
Drift is when the track slowly wanders off the plate over time. At frame 1 the track is solid; by frame 100 the CG element is six inches off where it should be; by frame 300 it’s noticeably wrong. The drift is gradual — no single frame looks dramatically wrong, but the cumulative error builds up.
What it looks like in playback: a CG element that starts in the right position relative to the plate but slowly slides as the shot progresses. The drift might be linear (constant rate) or it might accelerate.
Why drift happens: drift is almost always a calibration error somewhere in the camera solve. The focal length might be off by a few millimeters. The lens distortion model might be slightly wrong. The principal point (the optical center of the lens) might be assumed at the image center but actually offset. Each of these errors accumulates frame over frame as the camera moves through the scene.
How to diagnose: solve the track again with different starting conditions — different assumed focal length, different distortion parameters. If the drift changes character with the changes, the calibration is the problem. If the drift stays similar regardless of solve parameters, the issue is elsewhere (often plate quality, see Failure Mode 2).
How to fix: get better calibration data. If a lens distortion grid was shot, use it. If focal length and sensor size are documented in the camera report, use them. If the calibration data isn’t available, the matchmover has to reverse-engineer it from the plate, and the reverse-engineered values are educated guesses. Iterate the solve with different guesses until the drift goes away.
The frequency of drift in tracks is a good signal of how well the production captured camera data. Productions with strong on-set discipline have less drift in their tracks; productions that improvised camera reports have more.
Failure Mode 2: Slip
Slip is when the track loses sync with the plate by a few frames. The CG element is tracking correctly in space but lagging or leading the plate in time. The element appears to “skate” — moving in the right pattern but at a different timing than the plate.
What it looks like in playback: the CG element matches the plate’s motion in shape but is offset in time. A character moving from left to right in the plate has a CG hat that moves from left to right with them, but the hat starts moving a frame after the character does (or before). The disconnect is small but visible.
Why slip happens: slip is usually the result of bad tracking points being weighted too heavily in the solve. If the matchmove software is averaging across hundreds of tracked points, and a subset of those points have noisy or shifting positions, the solve compromises between the good points and the bad. The compromise produces a track that’s slightly out of sync.
Other causes: rolling-shutter artifacts in the plate (the bottom of each frame was captured a few milliseconds after the top, distorting the motion characteristics), interlaced footage being treated as progressive (every other field is offset), or motion blur being interpreted as actual subject motion by the tracker.
How to diagnose: examine the individual tracked points across frames. Points that drift in their position (the same physical feature in the scene moving from frame to frame) are bad points. Points that hold steady are good points. If the bad points outnumber the good, the solve is being polluted.
How to fix: prune the bad points and re-solve. The art is in identifying which points to keep — points on stable features (corners of architecture, distinctive marks on stationary surfaces) are more reliable than points on ambiguous features (flat surfaces, repeating patterns). Manual point selection by an experienced artist usually beats automatic point detection on difficult plates.
If the slip is from rolling shutter or interlacing, the fix is plate cleanup before the track. The shutter or interlacing has to be corrected at the plate level; the track can’t compensate for it through pure tracking work.
Failure Mode 3: Parallax Errors
Parallax is the apparent motion of objects at different distances from the camera as the camera moves. Near objects appear to move more in the frame than far objects do. A correctly tracked plate has its 3D scene reconstructed such that this parallax is preserved — near objects in CG move at the right rate, far objects at their right rate.
A parallax error is when the foreground tracks correctly but the background doesn’t (or vice versa). The CG element placed in the foreground sits beautifully in the scene; the CG element placed in the background drifts because the background’s depth wasn’t reconstructed correctly.
What it looks like in playback: a CG sky that doesn’t quite move with the rest of the scene as the camera pans. A CG building in the distance that wobbles when it should be solid. Foreground CG elements that look perfect; background CG elements that don’t quite settle.
Why parallax errors happen: parallax errors come from insufficient depth information in the solve. The matchmove software needs tracking points at multiple distances from the camera to reconstruct the scene’s depth. If most of the points are at one distance (all in the foreground, for example), the depth at other distances is underconstrained — the software is guessing.
How to diagnose: examine the reconstructed point cloud. A good 3D solve has tracked points distributed in depth — some near the camera, some at mid-distance, some far away. A bad solve has points clustered at one distance. The clustering is the problem.
How to fix: get more tracking points at the missing depths. If the plate has features that can be tracked at multiple distances, the artist can add manual tracks where the automatic detection didn’t. If the plate genuinely doesn’t have features at certain depths (a foreground subject against a featureless background, for example), the depth has to be supplied externally — through set survey or lidar data, through deliberately added tracking markers at known distances, or through manually informed depth assumptions.
In severe cases, the only fix is acknowledging that the plate can’t support the CG element you wanted to place there. A CG sky in a shot with no background tracking points to validate against may need to be a 2D solution instead of a tracked 3D one.
Diagnosis Before Intervention
The pattern that separates working matchmove artists from struggling ones is diagnosis-first thinking. When a track fails review, the impulse is to “improve the solve” — more tracking points, more iterations, different solver settings. Sometimes this works. Often it makes things worse, because the intervention doesn’t match the failure mode.
A drift problem with a slip-mode fix becomes worse drift. A slip problem with a parallax-mode fix becomes worse slip. Each failure mode has its own root cause, and the fix has to address the root cause directly.
The diagnosis takes 5–10 minutes per shot and saves 30–60 minutes of wrong-direction iteration. Across a project of hundreds of shots, the time savings is meaningful.
What the Best Plates Have
The plates that produce the cleanest tracks share a few characteristics:
Documented camera calibration. Focal length, sensor size, lens distortion data — all in the camera report, all delivered with the plate. This eliminates the calibration class of drift.
Multiple depths in the frame. Foreground elements, mid-ground elements, distant background elements. The depth distribution provides the points needed for accurate parallax reconstruction.
Stable, distinct features. Corners, edges, distinctive marks on solid surfaces. Avoid plates that are mostly flat textured surfaces (smooth walls, sand, snow) without tracking markers.
Reasonable shutter and motion characteristics. Standard shutter angles, no rolling-shutter artifacts, motion blur consistent with the frame rate. Plates with unusual capture parameters require additional cleanup.
Productions that capture plates with these characteristics have matchmove tracks that solve cleanly on the first attempt. Productions that don’t have tracks that take 3–5x longer per shot, with more failure modes to diagnose and fix.
How FXiation Digitals Approaches Track Failures
Our matchmove team diagnoses before intervening. Every track that doesn’t pass the first internal QC gets an explicit diagnosis — drift, slip, parallax, or a combination — before the fix work starts. The discipline takes a few extra minutes per shot but produces fewer failed iterations and faster overall delivery.
When we receive plates whose camera data is incomplete or whose features are insufficient for clean tracking, we’ll flag it before the work starts. Sometimes the plate can be tracked with extra effort; sometimes it requires production-side intervention (additional plate captures, reshoots, or methodology changes). We’d rather have the conversation upfront than discover a plate is untrackable mid-engagement.
If you’re scoping a project with significant tracking requirements, send us the camera reports and shot list and we’ll mark up which shots are likely to track cleanly, which need additional reference, and which have the kinds of capture characteristics that produce the failure modes above. The conversation is short and consistently saves multiples of the time it costs.
Common Questions