3D Model Alignment Using Geometric Feature Pair Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for aligning 3D models face challenges with planar or smooth surfaces, symmetric models with small asymmetric features, and require significant effort for training learning-based networks, leading to incorrect alignments and inefficiencies.
Innovation Solution
A method involving identifying geometric features in 3D models, assigning feature vectors, pairing them, calculating pair vectors, and determining a transformation matrix based on matching pairs, utilizing voxel-based parallel processing to enhance alignment accuracy and efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If point pair feature (PPF) method is used for aligning 3D models, then alignment can be performed on general surfaces, but alignment fails or becomes ambiguous for planar or smooth surfaces due to three ambiguous angles
Solution Approach 1:
The patent segments the 3D model into boundary points and interior points, using only boundary points for alignment. This segmentation resolves the ambiguity issue by focusing on the unique geometric characteristics of boundary points rather than using all points including interior points on planar surfaces.
Solution Approach 2:
The patent introduces asymmetric weighting for boundary points based on their local geometric properties. By assigning different weights to different boundary points according to their asymmetry metrics, the method resolves the three-angle ambiguity problem and achieves precise alignment even on planar surfaces.
2Measurement precision
If boundary point-to-boundary point (B2B) feature is used to solve planar surface alignment, then alignment works on planar surfaces, but wrong alignment occurs for symmetric models with small asymmetric features
Solution Approach 1:
The patent applies local quality by assigning different weights to different boundary points based on their local geometric properties. Points with higher asymmetry or more distinctive local features receive higher weights, ensuring that small asymmetric features in symmetric models are properly captured and used for correct alignment.
Solution Approach 2:
The patent changes the parameter weighting scheme from uniform to asymmetric based on local geometric properties. By dynamically adjusting the weight parameters for each boundary point according to metrics like curvature, edge sharpness, and local symmetry, the method reliably identifies and aligns small asymmetric features in symmetric models.
3Measurement precision
If learning-based weighting method is used to assign proper weights to points, then alignment accuracy improves for symmetric models, but significant effort and time are required to train the network
Solution Approach 1:
The patent implements self-service by automatically computing weights for boundary points based on their intrinsic geometric properties without requiring external training data or labeled poses. The weighting scheme is derived directly from the model's geometry through computational metrics, eliminating the need for time-consuming network training while achieving accurate alignment.
Data Source
AI summary
A computer-implemented method of aligning a source model with a target model includes receiving the source model and the target model, identifying geometric features in each of the source model and target model, assigning a feature vector to each feature, defining an associated geometry type, position, direction and magnitude for the feature, pairing each feature vector in the source model with each other feature vector in the source model and pairing each feature vector in the target model with each other feature vector in the target model, calculating a pair vector for each pairing, defining the geometry type of each feature vector in the pairing, the dimension of each feature vector in the pairing, a relative orientation and separation distance, identifying matching pair vectors between the source model and target model; and calculating a transformation matrix between the source model and target model based on the matching pair vectors.