A monocular depth prior based robust optimization method and system for sparse point clouds
By employing a robust optimization method for sparse point clouds based on monocular depth priors, and utilizing gradient-aware masks and dynamic anchor point mechanisms, the edge artifacts and scale displacement inconsistencies in point cloud reconstruction under sparse view scenarios are resolved, achieving efficient and robust point cloud optimization and accurate reconstruction.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- TAIYUAN UNIVERSITY OF TECHNOLOGY
- Filing Date
- 2026-02-27
- Publication Date
- 2026-07-17
AI Technical Summary
In sparse view or weak texture scenes, traditional SfM algorithm has difficulty generating high-quality initial point clouds. Monocular depth estimation suffers from edge artifacts and scale-displacement inconsistencies, leading to inaccurate point cloud reconstruction.
A robust optimization method for sparse point clouds based on monocular depth prior is adopted. A spatial confidence mask is constructed by calculating the gradient magnitude of the monocular depth map, and robust alignment and iterative optimization are performed. The confidence mask is used to shield edge noise, and the point cloud geometric coordinates are optimized by combining a multi-view reliability dynamic anchor point mechanism.
It effectively eliminates edge artifacts, solves the scale and displacement ambiguity problem between monocular depth and SfM point clouds, improves the geometric accuracy and structural stability of point clouds, and is computationally efficient and easy to integrate.
Smart Images
Figure CN122415398A_ABST