Research map / 3D vision / Human pose estimation
OrthoTrack: Continuous 6-DoF UAV Trajectory Estimation Anchored in Public Orthophotos
Oussema Dhaouadi, Zuria Bauer, Johannes Meier, Olaf Wysocki, Marc Pollefeys, Daniel Cremers
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 |
|---|---|---|---|
| combine keyframe localization with optical-flow tracking instead of localizing every frame (ablation) | OrthoTrack · MovingDrone (test) | ATE (m), FPS: ATE 4.83 -> 0.67; FPS 1.9 -> 23.8 | Improved |
| add periodic keyframe re-anchoring to flow tracking (ablation) | OrthoTrack · MovingDrone (test) | ATE (m): 27.70 -> 0.67 | Improved |
| add flow tracking to per-frame localization (ablation) | OrthoTrack · UAVScenes | TE vs ATE: TE 1.28 -> 1.38; ATE 1.89 -> 1.51 | Mixed |
| use learned optical flow (RAFT, NeuFlow v2) instead of Lucas-Kanade (ablation) | OrthoTrack · MovingDrone (test) | ATE (m): LK 0.67; RAFT 0.66; NeuFlow v2 0.82; SEA-RAFT 1.93 | No clear effect |
| use RoMa v2 Precise instead of weaker matchers as keyframe matcher (ablation) | OrthoTrack · MovingDrone (test) | ATE (m): 0.67 vs Base 0.90, Fast 0.88, GIM 2.56, L2M 6.23, MASt3R 5.74 | Improved |
| lower input frame rate from 30 fps to 2 fps (ablation) | OrthoTrack · MovingDrone (test) | ATE (m): 0.67 -> 0.71 (10 fps: 0.58) | No clear effect |
| use adaptive dual-threshold keyframe trigger instead of fixed reprojection threshold (ablation) | OrthoTrack · MovingDrone (test) | ATE (m): 0.79 -> 0.67 | Improved |
| use fixed keyframe interval K=50 instead of adaptive trigger (ablation) | OrthoTrack · MovingDrone (test) | ATE (m): 0.67 -> 0.53 at 30 fps; 12.26 at 2 fps | Mixed |
| use point-count-only keyframe trigger (ablation) | OrthoTrack · MovingDrone (test) | ATE (m): 0.67 -> 1.69 | Hurt |
| add oracle GNSS and magnetometer to piecewise VO alignment | DROID-SLAM · UAVScenes | RE (deg): RE 46.63 -> 35.02 | No clear effect |
| sweep 12 hyperparameters over wide ranges (ablation) | OrthoTrack · MovingDrone (test) | ATE (m): max deviation 0.26 m | No clear effect |
Closest papers by meaning
- OrthoLoC: UAV 6-DoF Localization and Calibration Using Orthographic Geodata (NeurIPS 2025 · similarity 0.91)
- ContinuLoc: Continuous Pose Inference over Neural Fields for UAV Geo-Localization (NeurIPS 2026 · similarity 0.87)
- Seeing Across Skies and Streets: Feedforward 3D Reconstruction from Satellite, Drone, and Ground Images (NeurIPS 2026 · similarity 0.87)
- Game4Loc: A UAV Geo-Localization Benchmark from Game Data (AAAI 2025 · similarity 0.85)
- View From Above: Orthogonal-View aware Cross-view Localization (CVPR 2024 · similarity 0.85)
- SatNav: A Scalable Benchmark for Long-Horizon UAV Vision-Language Navigation from Satellite Imagery (NeurIPS 2026 · similarity 0.84)
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.
