Three-dimensional reconstruction method, device, equipment, medium and product
By using 2D Gaussian sputtering technology and semantic mask supervision, adjusting 2D Gaussian primitives, and combining them with the TSDF algorithm, the problem of feature point matching difficulties on weak textures and reflective surfaces in traditional 3D reconstruction methods is solved, achieving efficient and accurate 3D mesh reconstruction. The generated model has clear edges and smooth textures.
Patent Information
- Application Number
- CN202610773968.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-06-01
- Publication Date
- 2026-07-17
AI Technical Summary
Existing 3D reconstruction methods face difficulties in feature point matching when dealing with weakly textured regions, reflective surfaces, and fine structures, leading to holes or geometric noise and poor reconstruction results. Furthermore, traditional methods are cumbersome and difficult to optimize through end-to-end error backpropagation.
A two-dimensional Gaussian sputtering technique is used to degenerate a three-dimensional ellipsoid into a two-dimensional flat disk that fits the surface of the object. The loss value is calculated by semantic masking and cumulative opacity. The two-dimensional Gaussian elements are adjusted until the iteration stopping condition is met. The three-dimensional mesh model is extracted by combining the TSDF algorithm.
It improves the 3D reconstruction effect, and the generated mesh model has clean and sharp edges, requiring no subsequent trimming or Laplacian smoothing. The texture is clear and the edges are smooth and continuous, making it suitable for industrial-grade modeling.
Smart Images
Figure CN122415947A_ABST