3D Morphable Model Registration via Surface Flattening
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for generating three-dimensional morphable models face challenges in registering points across training examples, particularly for complex shapes like ears, due to sensitivity to initialization, occlusions, and loss of semantic information, and are limited by the need for texture information and convex shape assumptions.
Innovation Solution
A method that determines shape parameters at each point on a three-dimensional meshed surface, flattens the surface to a two-dimensional representation, registers points based on these parameters, and down-samples the mesh to create a model with an average shape and deformation modes, allowing for local curvature and shape descriptors to adapt to the object's complexity, and uses techniques like ABF or LSCM for flattening and segmentation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If optical flow algorithm is used for point registration, then registration can be performed on training examples, but the method is very sensitive to initialization and requires small deformations between examples
Solution Approach 1:
The patent applies preliminary flattening transformations to map 3D mesh surfaces onto 2D planes before registration. This preliminary action creates a standardized 2D representation that eliminates the sensitivity to initialization and large deformations, allowing direct application of 2D image registration algorithms.
2Ease of operation
If cylindrical representation is used for 3D data, then 2D texture image is available for processing, but occlusions are generated and information is lost
Solution Approach 1:
The patent transforms the problem from 3D space to 2D space by flattening mesh surfaces onto 2D planes. This dimensionality change allows the use of成熟的2D image processing algorithms while preserving all surface information without occlusions, as the flattening process unwraps the surface completely.
3Measurement precision
If dense registration of several thousand points is performed, then sufficient association is achieved, but the complexity and computational cost increase significantly
Solution Approach 1:
The patent creates a simplified 2D copy or representation of the 3D mesh surface through flattening. This 2D copy retains all necessary geometric information for registration but is much simpler to process, allowing efficient application of standard 2D image registration techniques without the complexity of 3D point cloud registration.
4Loss of information
If manual registration of characteristic points is performed, then semantic information is preserved, but human resource costs increase
Solution Approach 1:
The patent enables the system to automatically perform registration by flattening surfaces and applying 2D image registration algorithms. This self-service approach eliminates the need for manual characteristic point registration while preserving semantic information through the automatic detection and matching of corresponding features on flattened surfaces.
Data Source
AI summary
A method is provided for generating a three-dimensional morphable model of an element from an initial database of examples of such elements providing data allowing, for each of the elements of the initial database, a three-dimensional meshed surface based on points and on a triangular network connecting the points to be determined.