一种基于不确定性引导的鲁棒多视图立体匹配方法
By constructing a probabilistic depth estimation model and sparse monocular guided feature fusion, the problems of matching ambiguity in weak texture regions and semantic feature-driven gradient updates are solved, achieving higher accuracy and robustness in 3D reconstruction.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NORTHWESTERN POLYTECHNICAL UNIV
- Filing Date
- 2026-03-18
- Publication Date
- 2026-07-17
AI Technical Summary
Existing technologies suffer from severe matching ambiguity and depth artifacts when dealing with weakly textured regions and non-Lambertian surfaces. Furthermore, the semantic features-driven gradient updates lead to shortcut learning, resulting in a decrease in the model's generalization ability in practical applications.
By constructing a probabilistic depth estimation model, utilizing uncertainty-guided adaptive cascade operations and sparse monocular guided feature fusion, the depth search range is optimized. Semantic feature correction is introduced in the geometrically ambiguous region, and semantic interaction is restricted in the reliable region to generate the final depth map.
It significantly improves matching accuracy and cross-domain generalization ability in weakly textured regions, avoids semantic features dominating gradient updates, and improves the robustness and accuracy of 3D reconstruction.
Smart Images
Figure CN122415689A_ABST