Robust video tracking under rapid motion and severe occlusion

Develop robust video tracking methods that remain reliable in scenarios involving rapid motions or severe occlusions, thereby providing dependable trajectory information for sketch animation generation based on CoTracker-tracked points.

Background

The HyperSketch framework uses trajectory consistency loss based on tracking information extracted by CoTracker to maintain temporally coherent vector sketch strokes. The paper notes that tracking quality can deteriorate when objects move rapidly or become severely occluded, which directly undermines the reliability of the trajectory constraints used during optimization.

Although the authors experimented with alternative trackers, including SpaTrackerV2 and Shape of Motion, they identify robust tracking in these challenging video conditions as an unresolved problem. Solving it would improve the temporal coherence and applicability of controllable video sketch generation in difficult scenes.