3D Model Reconstruction Using Multi-Resolution Image Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional 3D reconstruction methods face challenges in achieving accurate and dense 3D models due to low resolution images, resulting in noisy point clouds and long processing times, while using high-resolution images improves density but increases noise and processing time.
Innovation Solution
The method involves down-sampling 2D images to create low-resolution images for initial 3D point cloud reconstruction, followed by increasing 3D points using high-resolution images to improve accuracy and density, allowing for efficient reconstruction with reduced noise and processing time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If high-resolution images are used for 3D reconstruction, then the density and accuracy of the 3D model is improved, but the processing time and computational load increase significantly
Solution Approach 1:
The patent segments the 3D reconstruction process into multiple stages: initial reconstruction using low-resolution images to establish a coarse 3D model, followed by progressive refinement using progressively higher resolution images. This segmentation allows the system to benefit from both low-resolution speed and high-resolution accuracy at different stages of the process.
Solution Approach 2:
The patent performs preliminary 3D reconstruction using down-sampled low-resolution images before processing high-resolution images. This preliminary action establishes a initial 3D structure and point cloud that guides subsequent high-resolution processing, reducing the overall computational burden while maintaining final accuracy.
2Quantity of substance
If high-resolution images are used for 3D reconstruction, then the density of the point cloud is improved, but the noise in the point cloud increases
Solution Approach 1:
The patent divides the point cloud generation into multiple passes: an initial pass using low-resolution images to generate a sparse but clean point cloud, followed by subsequent passes using higher resolution images to progressively densify the point cloud while maintaining noise control through the staged approach.
Solution Approach 2:
The patent performs preliminary processing of high-resolution images including down-sampling and noise filtering before using them for point cloud generation. This preliminary action removes harmful noise components while preserving essential structural information needed for accurate 3D reconstruction.
3Productivity
If low-resolution images are used for 3D reconstruction, then the processing time is reduced, but the accuracy and density of the 3D model deteriorates
Solution Approach 1:
The patent segments the reconstruction workflow into multiple resolution levels, processing images at different resolutions in sequence. Low-resolution images are processed first for rapid initial model creation, then mid-resolution and high-resolution images are processed in subsequent stages to progressively improve accuracy, combining the speed benefits of low-resolution processing with the accuracy benefits of high-resolution processing.
Solution Approach 2:
The patent uses low-resolution images for preliminary reconstruction to establish the basic 3D structure quickly, then uses this preliminary model as a foundation for subsequent high-resolution processing. This preliminary action at low resolution provides a head start that reduces the overall processing time while the final high-resolution processing ensures accuracy.
Data Source
AI summary
A generation method is disclosed. Two-dimensional (2D) images that are generated by photographing a space from different viewpoints with at least one camera are obtained. Resolutions of the 2D images are reduced to generate first images, respectively. Second images are generated from the 2D images, respectively such that a resolution of each of the second images is higher than a resolution of any one of the first images. First three-dimensional (3D) points are generated based on the first images. The first 3D points indicate respective first positions in the space. A second 3D point is generated based on the second images. The second 3D point indicates a second position in the space. A 3D model of the space is generated based on the first 3D points and the second 3D point.