Guided Sparse Feature Matching via Coarse Dense Matches
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional feature matching techniques face challenges in achieving quarter-pixel accuracy while being computationally efficient, as dense matching is computationally heavy and lacks spatial smoothness, while sparse matching has a high rate of mismatches due to the absence of geometrical and spatial constraints.
Innovation Solution
The method employs guided sparse feature matching using coarsely defined dense matches, which derive semi-dense feature correspondences with better than quarter-pixel accuracy by incorporating spatial smoothness and geometric constraints, reducing computational cost and outlier rates through a multi-layer refinement stage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If dense feature matching is used to achieve quarter-pixel accuracy, then measurement precision is improved, but computational cost increases significantly
Solution Approach 1:
The patent segments the feature matching process into two distinct stages: a coarse dense matching stage that operates on down-sampled images to establish initial correspondences, and a fine sparse matching stage that refines these correspondences on full-resolution images. This segmentation allows the system to benefit from the accuracy of dense matching while avoiding its computational burden by limiting dense operations to down-sampled images.
Solution Approach 2:
The patent performs preliminary dense feature matching on down-sampled images before conducting the final sparse matching on full-resolution images. This preliminary action establishes a set of candidate correspondences that guide the subsequent refinement process, ensuring that the final matching operates on a constrained set of promising candidates rather than all possible feature pairs.
2Productivity
If sparse feature matching is used to reduce computational cost, then productivity is improved, but measurement precision deteriorates due to high mismatch rate
Solution Approach 1:
The patent introduces down-sampled images as an intermediary medium between the coarse dense matching and fine sparse matching stages. These down-sampled images provide a computational bridge that allows the system to establish initial correspondences efficiently while maintaining a pathway to high-precision refinement on full-resolution images, thus mediating between computational efficiency and accuracy requirements.
Solution Approach 2:
The patent replaces the traditional single-stage sparse matching mechanism with a two-stage process that incorporates dense matching on down-sampled images. This substitution transforms the mechanical approach from direct sparse matching to a guided process where dense correspondences on coarse images inform and constrain the sparse matching on fine images, reducing mismatches while maintaining efficiency.
3Device complexity
If conventional sparse matching is used without geometrical constraints, then device complexity is reduced, but reliability deteriorates due to absence of spatial smoothness
Solution Approach 1:
The patent adds a temporal dimension to the matching process by performing matching at multiple resolution levels (down-sampled and full-resolution images). This multi-scale approach introduces a new dimension of control where correspondences are first established on coarse scales and then refined on fine scales, enabling the incorporation of geometric constraints without overwhelming complexity at any single level.
Data Source
AI summary
An example method is described herein. The method includes executing dense feature matching on an image pair that is down sampled to obtain a first set of feature correspondences for each pixel of the down sampled image pair. The method also includes calculating a neighborhood correspondence based on the first set of feature correspondences for each pixel in a first image of the image pair. Further, the method includes executing sparse feature matching on stereoscopic patch pairs from the image pair based on the neighborhood correspondence for each pixel to obtain correspondence estimates for each stereoscopic patch pair. Finally, the method includes refining the correspondence estimates for each stereoscopic patch pair to obtain a semi-dense set of feature correspondences by applying a geometric constraint to the correspondence estimates and retaining correspondences that satisfy the geometric constraint.