一种基于二维-三维匹配的空间非合作目标位姿估计方法
By constructing a spatial non-cooperative target pose estimation method based on 2D-3D feature matching, a 3D feature reference model is reconstructed using multi-view images. Combined with a coarse-to-fine feature matching strategy using self-attention and cross-attention mechanisms, the stability problem of feature matching in complex environments of existing methods is solved, achieving high-precision and robust pose estimation.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HUNAN UNIV
- Filing Date
- 2026-04-14
- Publication Date
- 2026-07-17
AI Technical Summary
Existing spatial non-cooperative target pose estimation methods suffer from poor feature matching stability under conditions of significant changes in illumination, viewpoint, or insufficient target surface texture. Furthermore, they are highly dependent on target category and imaging conditions, resulting in limited generalization ability.
A two-dimensional-to-three-dimensional matching method is adopted to reconstruct a three-dimensional feature reference model through multi-view images. A coarse-to-fine feature matching strategy combining self-attention and cross-attention mechanisms is used, and a differentiable reprojection error optimization model is used for pose estimation.
It achieves high-precision and robust pose estimation under complex lighting and partial occlusion conditions, without requiring an accurate prior CAD model of the target, thus improving the accuracy and robustness of feature matching.
Smart Images
Figure CN122023535B_ABST