Spanning Tree Stereo Matching for Efficient Disparity Calculation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current stereo matching methods require complex optimizations and large data sets, especially when propagating additional information from ground control points, which can lead to inefficiencies in calculating disparity and depth information across entire images.

Innovation Solution

The method employs a spanning tree structure to propagate additional information from reference pixels to general pixels, using edge costs and data costs based on brightness and disparity information, allowing for efficient calculation of accumulated data costs and improved stereo matching and up-sampling without the need for extensive data sets.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If complex optimization methods are used to propagate additional information from ground control points, then the quality of stereo matching may be improved, but the computational complexity and time consumption increase significantly

Engineering Contradiction:
Improvestereo matching qualityVSAvoidcomputational complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent segments the image into multiple regions based on depth information, processing each region separately with appropriate methods. This divides the complex global optimization problem into smaller, more manageable local problems, reducing overall computational complexity while maintaining matching quality in each region

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent performs preliminary depth estimation using coarse-to-fine methods before conducting detailed stereo matching. This preliminary action provides initial guidance that simplifies subsequent optimization steps, reducing the search space and computational burden of the full matching process

Inventive Principle:
Principle #10Preliminary action

2Measurement precision

If large data sets are used for all pixels of the first image, then the accuracy of disparity calculation may be improved, but the memory requirements and processing time increase

Engineering Contradiction:
Improvedisparity accuracyVSAvoiddata set size
Core Design Contradiction:
Measurement precisionVSQuantity of substance

Solution Approach 1:

The patent applies different processing strategies to different pixels based on their local characteristics. Pixels in reliable depth regions use simplified methods with smaller data sets, while pixels in uncertain regions receive more intensive processing. This local differentiation maintains overall accuracy while reducing total data requirements

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The patent enables pixels to determine their own processing needs based on local depth reliability metrics. Each pixel region autonomously selects appropriate matching strategies and data requirements, eliminating the need for uniform large-scale data processing across the entire image

Inventive Principle:
Principle #25Self-service

3Ease of manufacture

If traditional cost aggregation methods are used without spanning tree structure, then the implementation is simpler, but the propagation of additional information from reference pixels is less efficient

Engineering Contradiction:
Improveimplementation simplicityVSAvoidinformation propagation efficiency
Core Design Contradiction:
Ease of manufactureVSProductivity

Solution Approach 1:

The patent introduces a spanning tree structure as an intermediary framework that connects reference pixels to general pixels through optimal paths. This intermediary structure efficiently guides the propagation of additional information while maintaining a relatively simple implementation through standard tree traversal algorithms

Inventive Principle:
Principle #24Intermediary (Mediator)

Data Source

PatentEP2947626B1Method and apparatus for generating spanning tree, method and apparatus for stereo matching, method and apparatus for up-sampling, and method and apparatus for generating reference pixel
Publication Date: 2020.04.08 SAMSUNG ELECTRONICS CO LTD
  • EP2947626B1 patent drawingFigure 1
  • EP2947626B1 patent drawingFigure 2
  • EP2947626B1 patent drawingFigure 3

AI summary

A method and apparatus for generating a spanning tree, a method and apparatus for stereo matching, a method and apparatus for up-sampling, and a method and apparatus for generating a reference pixel are disclosed, in which a spanning tree may be generated by reference pixels, stereo matching or up-sampling may be performed based on the generated spanning tree, and a reference pixel may be generated based on a stereo video.