Leveraging AV1 motion vectors for Fast and Dense Feature Matching
About
We repurpose AV1 motion vectors to produce dense sub-pixel correspondences and short tracks filtered by cosine consistency. On short videos, this compressed-domain front end runs comparably to sequential SIFT while using far less CPU, and yields denser matches with competitive pairwise geometry. As a small SfM demo on a 117-frame clip, MV matches register all images and reconstruct 0.46-0.62M points at 0.51-0.53,px reprojection error; BA time grows with match density. These results show compressed-domain correspondences are a practical, resource-efficient front end with clear paths to scaling in full pipelines.
Julien Zouein, Hossein Javidnia, Fran\c{c}ois Piti\'e, Anil Kokaram• 2025
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Structure-from-Motion | Paris Seq. 1 | Registered Images117 | 5 | |
| Feature Matching | Dublin Seq 1 | Median Inlier Ratio99 | 4 | |
| Feature Matching | Paris Seq. 1 | Median Inlier Ratio99 | 4 | |
| Feature Matching | Paris Seq 2 | Median Inlier Ratio99 | 4 | |
| Feature Matching | KITTI Seq 0 | Median Inlier Ratio98 | 4 | |
| Feature Matching | Gerrard Hall | Median Inlier Ratio96 | 4 | |
| Geometric Verification | Dublin Seq 1 | Median Sampson Error3.00e-4 | 4 | |
| Structure from Motion efficiency | Dublin Seq 1, Paris Seq 1, Paris Seq 2, KITTI Seq 0, Gerrard Hall, and Person Hall (Median over all sequences) | Pre-Processing Time (s)12.23 | 4 | |
| Feature Matching | Person Hall | Median Inlier Ratio96 | 4 | |
| Geometric Verification | Paris Seq. 1 | Median Sampson Error5.82e-5 | 3 |
Showing 10 of 14 rows