A multi-focus fusion DOF three-dimensional reconstruction method based on a Siamese network
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- BEIJING BOVISION TECH CO LTD
- Filing Date
- 2026-05-09
- Publication Date
- 2026-07-24
AI Technical Summary
Existing DOF 3D reconstruction methods suffer from unstable focus area judgment under complex imaging conditions, leading to misjudgment of local pixels and affecting the continuity and accuracy of depth reconstruction results.
The Siamese network is used for multi-focal plane image feature extraction. Focus feature maps are generated through CNN, FPN and DCN. Focus decision maps and confidence maps are generated by combining focal plane depth, threshold and weight parameters. Pixel-level fusion and index matrix update are performed to correct abnormal index jumps and generate optimized 3D reconstruction results.
It improves the ability to identify focusing misjudgments in areas with weak texture, high reflectivity, and noise, enhances the continuity and reliability of 3D reconstruction results, and reduces the impact of depth abrupt changes and local errors.
Smart Images

Figure CN122454053A_ABST