Research map / 3D vision / Human pose estimation
Combining 3D Model Contour Energy and Keypoints for Object Tracking
Bogdan Bugaev, Anton Kryshchenko, Roman Belov
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add contour energy refinement after the KLT tracker (ablation) | KLT keypoint tracker · OPT benchmark (all tests) | AUC: AUC increased by 18% | Improved |
| add contour energy refinement after the KLT tracker (ablation) | KLT keypoint tracker · OPT benchmark (max velocity tests) | AUC: increase was 27% | Improved |
| combine keypoints and contour energy instead of ORB-SLAM2 | tracking · OPT benchmark (all tests) | AUC: 12.97 -> 14.79 | Improved |
| combine keypoints and contour energy instead of GOS edge-based tracking | tracking · OPT benchmark (all tests) | AUC: 7.68 -> 14.79 | Improved |
| use the proposed method instead of PWP3D | tracking · OPT benchmark (all tests) | AUC: 5.01 -> 14.79 | Improved |
| limit pose search area using keypoint reprojection error (ablation) | contour energy optimization · OPT benchmark (Soda, symmetrical object) | AUC: 3.12 -> 14.85 | Improved |
| use the proposed method instead of ORB-SLAM2 on z translation | tracking · OPT benchmark (z translation) | AUC: 7.07 -> 13.08 | Improved |
| use the proposed method instead of ORB-SLAM2 on flashlight | tracking · OPT benchmark (Flashlight) | AUC: 15.91 -> 13.96 | Hurt |
| use the proposed method instead of ORB-SLAM2 on House | tracking · OPT benchmark (House) | AUC: 17.28 -> 14.48 | Hurt |
| use the proposed method instead of ORB-SLAM2 on Chest | tracking · OPT benchmark (Chest) | AUC: 15.53 -> 14.97 | No clear effect |
| use the proposed method for speed vs GOS | tracking · OPT benchmark frame processing time | ms per frame: 201 -> 683 | Hurt |
| use larger 3D model (more faces) | proposed method · OPT frame processing time by object | ms per frame: 358 -> 1097 | Hurt |
Closest papers by meaning
- SpatialTrackerV2: Advancing 3D Point Tracking with Explicit Camera Motion (ICCV 2025 · similarity 0.87)
- Detect-and-Track: Efficient Pose Estimation in Videos (CVPR 2018 · similarity 0.86)
- Prior-free 3D Object Tracking (CVPR 2025 · similarity 0.86)
- Tracking by 3D Model Estimation of Unknown Objects in Videos (ICCV 2023 · similarity 0.86)
- Generalizable Structure-Aware Keypoint Correspondence for Category-Unified 3D Single Object Tracking (CVPR 2026 · similarity 0.86)
- KPA-Tracker: Towards Robust and Real-Time Category-Level Articulated Object 6D Pose Tracking (AAAI 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
