A computer vision tracking method for lock focus dynamic pull-up
By constructing an initial tracking point matrix, multi-scale structural tensor search, and Delaunay triangulation, combined with weighted averaging and principal component analysis, the feature degradation and positioning drift problems of target tracking in dynamic large lifting scenarios are solved, achieving high-precision target locking and stable tracking.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHENZHEN YIXING MEDICAL BEAUTY HOSPITAL
- Filing Date
- 2026-03-06
- Publication Date
- 2026-07-17
AI Technical Summary
Existing technologies are prone to feature degradation when dealing with rapid scaling and complex deformation of the target scale, making it difficult to adapt to instantaneous changes in the target's pose. This results in lag in the tracking bounding box response or positioning drift, making it impossible to accurately reconstruct the outline of the target body.
An initial tracking point matrix is constructed in the initial frame of the video sequence. Candidate matching points are searched through multi-scale structural tensors. Outliers are eliminated by Delaunay triangulation. The target center and principal axis direction are calculated by combining weighted average and principal component analysis. A bounding box with dynamic buffer spacing is generated, and the local search window is adaptively adjusted.
It significantly improves tracking robustness and accuracy in dynamic high-amplitude scenarios, effectively addressing the coupling challenge of rapid camera zoom and violent target movement, and achieving stable target locking and accurate tracking.
Smart Images
Figure CN122415673A_ABST