3D Footwear Mesh Alignment Using Proxy Foot Geometry
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for aligning and annotating items in augmented reality or virtual-try-on processes, such as footwear, are inaccurate and time-consuming, often requiring manual input or machine learning with significant computational costs.
Innovation Solution
Automatically determining location data and deforming object shapes using geometric mesh processing, without manual input or specialized hardware, by generating a coarser mesh with ambient occlusion and alignment to a proxy object shape, such as a human foot.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual or machine-learned processes are used to annotate locations in item geometry, then alignment accuracy between item and user body can be improved, but computational cost and processing time increase significantly
Solution Approach 1:
The patent uses a proxy mesh (copy) of the target body part to establish initial alignment and correspondence. Instead of performing complex machine learning annotation directly on the target image, the system creates a simplified proxy representation that can be aligned more efficiently, then transfers the alignment information back to the original geometry.
Solution Approach 2:
The patent introduces an intermediary alignment process using proxy meshes and correspondence points as a mediator between the item geometry and the target body part. This intermediary step simplifies the complex alignment problem into manageable transformations that can be computed more efficiently than direct annotation methods.
2Measurement precision
If manual processes are used to annotate item geometry, then alignment accuracy can be improved, but processing time and productivity decrease
Solution Approach 1:
The system creates a proxy mesh copy of the body part to perform alignment operations. This copy can be processed quickly without affecting the original high-resolution geometry, enabling faster iteration and processing while maintaining alignment accuracy when transferred back to the item geometry.
Solution Approach 2:
The patent segments the alignment process into distinct steps: creating proxy mesh, establishing correspondence points, computing transformations, and applying to item geometry. This segmentation allows each step to be optimized independently and processed in parallel where possible, improving overall processing speed.
3Measurement precision
If complex machine-learned annotation processes are used, then measurement precision can be improved, but device complexity and computational resources required increase
Solution Approach 1:
The patent replaces complex machine learning annotation with a simpler proxy-based approach. The proxy mesh serves as a simplified representation that captures essential geometric features without requiring complex ML models, reducing system complexity while maintaining sufficient annotation accuracy for alignment purposes.
Solution Approach 2:
The proxy mesh is a simplified, computationally inexpensive representation that can be created and discarded quickly. Instead of investing significant computational resources in complex ML annotation, the system uses this lightweight proxy approach that requires minimal computational resources while achieving the necessary alignment accuracy.
Data Source
AI summary
The shape of an article of footwear is initially represented by first mesh data. Second mesh data having fewer faces is determined by using an occlusion algorithm to determine occluded faces of the first mesh, using a first function to determine values based on the non-occluded faces, and a second function to determine values based on occluded and non-occluded faces. Using these values, the second mesh is spaced a minimum offset distance to enclose the first mesh without intersection. Proxy data representing the shape of a foot is aligned with the inner surface of the second mesh, then deformed to place the shape adjacent to the inner surface. The mesh representing the footwear may be deformed to place portions of the representation of footwear adjacent to the shape of the foot. Locations of keypoints in the shape of the foot are used to add keypoints to the mesh.