3D Scene Model Reconstruction from Image Subset Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for determining three-dimensional scene models from digital images face scalability issues and inaccuracies due to redundancy and noise in depth maps, particularly when processing consumer videos captured by hand-held cameras, which results in degraded image quality and inability to accurately synthesize new views.
Innovation Solution
A method that selects a subset of digital images with overlapping content and reduces redundancy by determining target camera positions and eliminating low-quality images, using a data processing system to analyze these images and construct a three-dimensional model efficiently.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If all digital images in the collection are processed to generate 3D model, then the completeness of scene coverage is improved, but the computational complexity and processing time increase significantly
Solution Approach 1:
The patent extracts and processes only a selected subset of digital images that contain valuable new information, rather than processing all images in the collection. This is achieved by identifying and eliminating redundant images through comparison of extracted features, thereby reducing computational complexity while maintaining scene coverage completeness.
Solution Approach 2:
The patent applies different processing strategies to different subsets of images based on their redundancy characteristics. High-quality non-redundant images are processed in detail, while redundant images are identified and excluded, creating a locally optimized processing approach that balances completeness and computational efficiency.
2Manufacturing precision
If depth maps are calculated for each image and merged, then the 3D reconstruction density is improved, but the noise and redundancy increase leading to degraded image quality
Solution Approach 1:
The patent performs preliminary selection and filtering of images before depth map calculation. By pre-identifying and eliminating redundant images through feature extraction and comparison, the system avoids calculating depth maps for unnecessary images, thereby reducing noise and redundancy in the final 3D reconstruction while maintaining adequate density.
Solution Approach 2:
The patent converts the potentially harmful effect of image redundancy into a beneficial filtering process. By systematically identifying redundant images through feature comparison, the method eliminates noise sources while preserving valuable overlapping information that enhances 3D reconstruction density, thus improving overall image quality.
3Ease of manufacture
If consumer videos from hand-held cameras are used, then the ease of data acquisition is improved, but the disordered camera trajectories and redundant frames reduce processing efficiency
Solution Approach 1:
The patent segments the video data into individual frames and processes them through a systematic filtering approach. By dividing the video sequence into discrete frames and applying redundancy elimination to each, the method maintains ease of data acquisition from hand-held cameras while improving processing efficiency through structured analysis.
Solution Approach 2:
The patent changes the processing parameters by selectively including or excluding frames based on redundancy analysis. By adjusting which frames are processed (changing the effective sample size and composition), the system maintains compatibility with easily acquired consumer video data while significantly improving processing efficiency through intelligent parameter selection.
Data Source
AI summary
A method for determining a three-dimensional model of a scene from a collection of digital images, wherein the collection includes a plurality of digital images captured from a variety of camera positions. A set of the digital images from the collection are selected, wherein each digital image contains overlapping scene content with at least one other digital image in the set of digital images, and wherein the set of digital images overlap to cover a contiguous portion of the scene. Pairs of digital images from the set of digital images to determine a camera position for each digital image. A set of target camera positions is determined to provide a set of target digital images having at least a target level of overlapping scene content. The target digital images are analyzed using a three-dimensional reconstruction process to determine a three-dimensional model of the scene.