3D Mesh Normal Selection via Co-Visibility
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current systems for creating three-dimensional mesh models of structures, such as landmarks, face errors due to incorrect normal fitting and directional ambiguity, especially when dealing with complex geometries like the Arc de Triomphe, leading to inaccurate mesh creation.
Innovation Solution
The method involves accessing a set of three-dimensional points from images, determining a reference image, identifying co-visible points within a visibility cone, and selecting a normal using these points to compute a three-dimensional mesh model, which improves mesh quality and reduces model size, enabling efficient navigation and display.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If normals are recovered by fitting a plane to a point and its nearest neighbors, then the mesh can be created, but the normals may be incorrect for complex geometries leading to large errors in mesh creation
Solution Approach 1:
The patent extracts the harmful assumption that nearest neighbors belong to the same surface. By removing this incorrect constraint and replacing it with co-visibility-based point selection, the system avoids fitting planes to geometrically inconsistent point sets, thereby eliminating the source of normal direction errors while maintaining mesh creation capability
Solution Approach 2:
The patent introduces co-visibility relationships as an intermediary criterion to select points for normal fitting. Instead of directly using spatial proximity (nearest neighbors), the system uses the mediator concept of mutual visibility through camera observations to identify points that truly belong to the same surface, resolving the contradiction between mesh creation and normal accuracy
2Weight of moving object
If a sparse point cloud is used to create a mesh model, then the model size is reduced, but the mesh quality deteriorates due to insufficient point information
Solution Approach 1:
The patent applies local quality by treating different regions of the point cloud differently based on their co-visibility relationships. Instead of uniformly sampling or thinning the point cloud, the system identifies and preserves points that have strong co-visibility connections, ensuring that locally critical points for defining surface geometry are retained while allowing less critical points to be removed, thus maintaining mesh quality with reduced model size
Solution Approach 2:
The patent changes the parameter used for point selection from simple spatial distance to co-visibility probability. By transforming the selection criterion from a geometric parameter (distance) to an observational parameter (visibility through cameras), the system can more effectively identify and preserve points that are critical for accurate mesh reconstruction, achieving better mesh quality at reduced model size
Data Source
AI summary
A method for creating a three-dimensional mesh model of a structure includes accessing a set of three-dimensional points associated with a set of images of the structure. For each three-dimensional point in the set of three-dimensional points, the method determines a reference image, identifies a subset of images from the set of images of the structure taken within a distance from the reference image, determines a subset of three-dimensional points seen by the subset of images, filters the subset of three-dimensional points to retain only a set of co-visible points that lie in a visibility cone of the reference image, and selects a normal using the set of co-visible points. The three-dimensional mesh model of the structure is computed using the selected normal, and the model may be provided to a second computing device.