Neural Network 3D Object Reconstruction for Pattern Correspondence

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing 3D scanning technologies face challenges in determining the correspondence between image pixels and projection pattern elements, particularly with non-coded elements, leading to ambiguity in reconstructing object surfaces.

Innovation Solution

Utilizing a trained neural network to infer the correspondence between image pixels and projection pattern elements using simulated data, which includes projected patterns, object shapes, and corresponding element data to enhance the accuracy of 3D reconstruction.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional triangulation algorithms are used to determine correspondence between image pixels and projection pattern elements, then the system is simple to implement, but the resolution and completeness of 3D reconstructions deteriorate, especially for complex geometries

Engineering Contradiction:

Improveresolution and completeness of 3D reconstructionVSAvoidcomplexity of correspondence determination system

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent replaces traditional geometric triangulation algorithms with a neural network-based system to determine correspondence between image pixels and projection pattern elements. The neural network is trained on simulated data containing projection patterns, object shapes, and element correspondences, enabling it to accurately resolve ambiguities in structured light scanning, particularly for complex geometries where traditional methods fail.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent employs preliminary action by training the neural network in advance using extensively pre-generated simulated data that includes diverse projection patterns, object shapes, and corresponding element mappings. This pre-training enables the network to handle various scanning scenarios without requiring real-time complex calculations during actual 3D reconstruction.

Inventive Principle:

Principle #10Preliminary action

2Ease of manufacture

If simulated data is used to train the neural network, then the training can be performed offline without requiring real object scanning, but the data generation process becomes complex and computationally intensive

Engineering Contradiction:

Improveease of neural network trainingVSAvoidcomplexity of simulated data generation system

Core Design Contradiction:

Ease of manufactureVSDevice complexity

Solution Approach 1:

The patent creates virtual copies of real-world scanning scenarios through simulated data generation. The simulation system generates synthetic projection patterns, object surfaces, and corresponding element mappings that replicate actual scanning conditions without requiring physical objects or complex experimental setups. This copying approach enables comprehensive training data generation offline.

Inventive Principle:

Principle #26Copying

Data Source

PatentEP4204760B1Systems and methods of 3D object reconstruction using a neural network

Publication Date: 2025.09.24 ARTEC EURO S A R L

Figure 1A
Figure 1B
Figure 1C~1D

AI summary

In accordance with some embodiments, a method is provided for determining correspondence between a projection pattern and an image of the project ion pattern shone onto the surface of an object. The method includes obtaining an image of an object while a projection pattern is shone on the surface of the object. The method further includes, using a neural network to output a correspondence between respective pixels in the image and coordinates of the projection pattern. The method further includes, using the correspondence between respective pixels in the image and coordinates of the projection pattern, reconstructing a shape of the surface of the object.