Feature Matching Using Local Epipolar Search for Camera Arrays
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional feature matching methods for camera arrays are too slow and inaccurate, especially for mobile devices, due to their high computational load and high false match rates, which are exacerbated by the difficulty in obtaining precise epipolar lines in non-planar scenes.
Innovation Solution
The implementation of a local epipolar-based search method that reduces the search area by using a general direction of the epipolar line, rather than a precise measurement, to efficiently match feature points between images in a camera array, thereby optimizing the feature matching process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional feature matching methods are used to match feature points across multiple camera images, then comprehensive feature point coverage is achieved, but the computational load increases and matching speed decreases
Solution Approach 1:
The patent divides the target image into multiple blocks and performs feature point matching only within localized search areas around projected feature points from the reference image, rather than searching the entire image. This segmentation approach reduces computational load while maintaining matching accuracy by concentrating search efforts in relevant regions.
Solution Approach 2:
The patent applies different search strategies to different regions by creating localized search areas around each projected feature point. Each search area is tailored to the specific geometric relationship between camera pairs, using local epipolar line directions rather than a uniform global search approach. This local adaptation improves both speed and accuracy.
2Measurement precision
If a large search area is used to ensure accurate feature point matching, then matching accuracy is improved, but the computational load and processing time increase
Solution Approach 1:
The patent performs preliminary projection of feature points from the reference image to the target image using estimated camera geometry, which defines constrained search areas before the actual matching process. This preliminary action eliminates the need for exhaustive full-image searches, reducing processing time while maintaining precision through targeted local searches.
Solution Approach 2:
The patent reduces the two-dimensional full image search space to one-dimensional local search areas defined by epipolar line directions. By constraining searches to lines rather than planes, and further to small regions around projected points, the search dimensionality is effectively reduced, dramatically cutting processing time while preserving matching precision.
3Measurement precision
If precise epipolar lines are obtained for non-planar scenes to improve matching accuracy, then feature matching precision is improved, but the system complexity and computational requirements increase
Solution Approach 1:
The patent uses approximate camera center positions and rough epipolar line directions as disposable, low-cost estimates rather than computing precise epipolar geometry. These simplified geometric models are sufficient to define adequate search areas for accurate matching without requiring complex scene understanding or precise calibration, reducing system complexity while maintaining practical accuracy.
Data Source
AI summary
A system, article, and method of feature matching for multiple images.