3D Reconstruction of Transparent Objects via Light Correspondence Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional three-dimensional reconstruction algorithms are ineffective for transparent objects due to the complexities of light refraction and reflection, which are not adequately addressed by existing methods.
Innovation Solution
A method involving silhouettes and light correspondences acquisition under different viewing directions, filtering out multiple refractions and reflections through backward ray tracing, and optimizing surface normal vectors and Snell normal vectors to create a point cloud reconstruction model, followed by Poisson sample point updates and silhouette consistency constraints for accurate three-dimensional reconstruction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional three-dimensional reconstruction algorithms are used, then the reconstruction process is simple, but they cannot properly reconstruct transparent objects due to light refraction and reflection complexities
Solution Approach 1:
The patent segments the complex light transport problem into distinct components: silhouette extraction, light correspondence identification, refraction/reflection filtering, and point cloud optimization. By dividing the reconstruction process into these manageable stages, the algorithm can systematically handle transparent objects while maintaining computational feasibility
Solution Approach 2:
The patent introduces light correspondences as an intermediary element that bridges the gap between silhouettes and three-dimensional reconstruction. By tracking how light rays correspond across different views and filtering out complex refraction/reflection effects, the method creates a reliable intermediate representation that enables accurate transparent object reconstruction
2Measurement precision
If light correspondences are fully utilized for reconstruction, then the surface restoration accuracy is improved, but the computational complexity increases due to filtering multiple refractions and reflections
Solution Approach 1:
The patent performs preliminary filtering of light correspondences to remove those involving multiple refractions and reflections before the main reconstruction process. By pre-processing and eliminating problematic light rays early in the pipeline, the method reduces computational burden on subsequent optimization stages while preserving the accuracy benefits of utilizing light correspondences
Solution Approach 2:
The patent optimizes the utilization of light correspondences by selectively processing only those that contribute meaningfully to surface reconstruction. Rather than exhaustively processing all possible light paths, the method focuses on the most informative correspondences, achieving high surface accuracy with reduced computational overhead
3Manufacturing precision
If backward ray tracing is used to filter light rays, then the quality of point cloud model is improved, but the processing time is extended
Solution Approach 1:
The patent extracts and removes light rays involved in multiple refractions and total reflections from the dataset before performing backward ray tracing. By eliminating these problematic rays in advance, the method reduces the computational scope of the ray tracing operation while maintaining the quality of the resulting point cloud model
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
Enables the reconstruction of transparent objects by fully utilizing light correspondences and optimizing the point cloud model, effectively restoring the entire surface of the object.
Implementation Method 1
filtering out light rays involved in multiple refractions and total reflections by backward ray tracing
Implementation Method 2
filtering out light rays involved in multiple refractions and total reflections by backward ray tracing
Implementation Method 3
performing a first optimization on the initial rough transparent object model by constraining consistency between the surface normal vectors and the Snell normal vectors
Data Source
AI summary
The present application relates to methods and apparatuses for three-dimensional reconstruction of a transparent object, computer devices, and storage mediums. The method includes acquiring silhouettes and light correspondences of a to-be-reconstructed transparent object under different viewing directions, acquiring an initial rough transparent object model of the to-be-reconstructed transparent object, and filtering out light rays involved in multiple refractions and total reflections by backward ray tracing; acquiring surface normal vectors and Snell normal vectors of the initial rough transparent object model in camera device views, and performing a first optimization on the initial rough transparent object model by constraining consistency between the surface normal vectors and the Snell normal vectors of the initial rough transparent object model in the camera device views, and acquiring a point cloud reconstruction model; acquiring Poisson sample points of the initial rough transparent object model, and updating the point cloud reconstruction model according to the Poisson sample points.