Image Sequence Subsetting for 3D Reconstruction Efficiency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current three-dimensional image reconstruction methods are computationally expensive and time-consuming due to the need for feature matching calculations across large numbers of two-dimensional images, which can be optimized by determining image correlations before performing feature matching.
Innovation Solution
Divide image sequences into subsets based on similarity, determining high correlation images within each subset to focus subsequent feature matching calculations only on these images, thereby reducing computational load and improving efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If feature matching calculation is performed on all two-dimensional images for three-dimensional image reconstruction, then reconstruction accuracy is improved, but computational cost and processing time increase significantly
Solution Approach 1:
The patent calculates correlation between image sequences before performing feature matching. By pre-determining which image sequences are highly correlated, the system can prioritize feature matching calculations on these sequences, avoiding unnecessary computations on unrelated images and thus reducing overall processing time while maintaining reconstruction accuracy.
Solution Approach 2:
The patent divides the set of two-dimensional images into multiple image sequences and further segments them based on correlation analysis. By organizing images into correlated groups rather than processing all images uniformly, the system reduces the computational burden while ensuring that feature matching is performed on the most relevant image pairs for accurate three-dimensional reconstruction.
2Measurement precision
If feature matching calculation is performed on all two-dimensional images, then reconstruction accuracy is improved, but computational resources are consumed excessively
Solution Approach 1:
The patent performs correlation calculation between image sequences as a preliminary step before feature matching. This pre-filtering identifies the most relevant image sequences that share high correlation, allowing the system to allocate computational resources only to these sequences for feature matching, thereby reducing overall resource consumption while maintaining the accuracy required for reliable three-dimensional reconstruction.
Solution Approach 2:
Instead of performing feature matching on all image pairs (excessive action), the patent applies partial action by selecting only the most correlated image sequences for feature matching. This selective approach ensures sufficient accuracy for reconstruction while significantly reducing computational resource consumption by avoiding unnecessary processing on uncorrelated images.
3Productivity
If correlation determination is performed before feature matching, then computational efficiency is improved, but additional calculation step is introduced
Solution Approach 1:
The patent introduces correlation determination as a preliminary step that organizes image sequences before feature matching. Although this adds an initial calculation step, it enables subsequent feature matching to be performed only on highly correlated images, resulting in significant overall efficiency gains that outweigh the additional complexity of the pre-processing stage.
Solution Approach 2:
The patent segments the image processing pipeline into distinct stages: correlation determination and feature matching. By separating these functions and using correlation results to guide feature matching selection, the system manages complexity through structured organization while achieving improved computational efficiency through targeted processing.
Data Source
Figure 1
Figure 2
Figure 3A~3B
AI summary
A method, an apparatus, a device, and a storage media for image processing are disclosed in the present disclosure and relate to the field of computer vision and three-dimensional image reconstruction. Detailed implemented solution includes a method of image processing, including: acquiring a set of image sequences, the set of image sequences including a plurality of image sequences; determining a first similarity measurement between image sequences in the set of image sequences; dividing the set of image sequences into one or more subset of image sequences based on a first similarity measurement; and determining, in each subset of image sequences, degrees of correlation between images in one image sequence of the subset of image sequences and images in other image sequences of the subset of image sequences. According to the embodiments of the present disclosure, it is possible to reduce the amount of calculation in three-dimensional image reconstruction, improve the calculation efficiency, and applicable to the field of automatic-driving.