Motion Tracking

Motion tracking analyzes the movement of pixels, patterns, or features across sequential frames to derive position, rotation, scale, and sometimes perspective data. There are three fundamentally different approaches, and using the wrong one is the most common tracking mistake. Point tracking (used in DaVinci Resolve's Tracker node, After Effects' built-in tracker) follows a small rectangular region — a user-defined search area around a feature with high contrast. It works by comparing the pixel pattern in the search region frame-to-frame, finding the best match, and recording the offset. Point tracking is fast and sufficient for simple stabilization or attaching a logo to a flat surface. It fails when the tracked point rotates significantly (the pattern changes shape beyond the tracker's ability to match), gets occluded (something passes in front of it), or leaves the frame. Planar tracking (Mocha Pro, Resolve's Planar Tracker, After Effects with Mocha AE) doesn't track a point — it tracks a surface. You define a region (a plane), and the algorithm uses the entire pattern within that region to calculate affine or perspective transforms. This is dramatically more robust than point tracking for surfaces that rotate, tilt, or change perspective. Want to replace a phone screen that rotates 45 degrees and partially goes off-screen? Planar track. The Mocha Pro plugin for After Effects and Nuke is the industry standard — its surface tracking handles reflections, partial occlusion, and out-of-frame surfaces better than any alternative. 3D camera tracking (After Effects' 3D Camera Tracker, Nuke's CameraTracker, PFTrack, SynthEyes) is a completely different beast. Instead of tracking a specific point or surface, it analyzes the parallax of features across the entire frame to reverse-engineer the 3D camera position, orientation, and lens properties for every frame. This gives you a virtual camera in 3D space that matches the real camera's movement — you can place 3D objects, text, or effects into the scene with correct perspective and depth.

Tracking fails for predictable reasons, and most have workarounds. Motion blur is enemy number one — fast-moving objects create streaked pixels that the tracker can't match between frames. Solution: increase the tracker's search region to be larger than the blur trail, or manually keyframe the track points during blurry sections and let the tracker fill in the rest. Low contrast is the second failure mode. Trackers work by matching pixel patterns — if there's no pattern (a uniform wall, a blurry sky, a dark shadow), there's nothing to match. The fix is to find a nearby feature with contrast and offset the attachment point. Occlusion is the third — when something passes in front of your tracked point. Mocha Pro handles this with its occlusion layer system (you tell Mocha what's in front, and it ignores those pixels), but point trackers simply lose the track. The fix is to split the track into segments and use different track points for each segment, blending the results. Rolling shutter is a sneaky failure mode on CMOS cameras — fast pans create wobble and skew that point trackers interpret as false motion. Planar trackers handle this better because they model the distortion as a perspective change, but it still introduces drift over long tracks. The honest workflow for difficult tracks: start with the automated tracker, evaluate the result frame-by-frame, manually correct drift at keyframes, and accept that about 30% of tracks need manual cleanup on complex shots. No tracking algorithm — including the AI-powered ones in the latest Resolve and After Effects versions — is 100% reliable on footage with heavy motion blur, low contrast, and frequent occlusions. That's why experienced compositors still know how to hand-track.