3D Shape Determination Using Neural Network Pattern Recognition

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current methods for determining three-dimensional shapes of objects using pattern decoding are inefficient due to high computational requirements and the need for detailed recognition of figures, especially when dealing with repetitive patterns.

Innovation Solution

A method involving the projection of a figure pattern with a repeating region onto an object, using a trained neural network to recognize figures in the recorded image, which allows for rapid determination of the three-dimensional shape by leveraging area indices and relative positions, reducing computational effort through epipolar geometry and calibration information.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If a conventional pattern decoding method is used to determine three-dimensional shape, then the shape can be determined, but high computational effort and time are required

Engineering Contradiction:
Improvethree-dimensional shape determinationVSAvoidcomputational time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The figure pattern is divided into multiple areas, each containing a subset of figures. The neural network processes each area separately and independently, breaking down the complex task of recognizing all figures in the entire pattern into smaller, parallelizable sub-tasks. This segmentation enables faster processing while maintaining accurate three-dimensional shape determination.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

Instead of requiring the neural network to recognize all figures in the complete figure pattern, the method uses only a subset of figures from selected areas. This partial action approach provides sufficient information for accurate three-dimensional shape reconstruction while significantly reducing computational complexity and processing time.

Inventive Principle:
Principle #16Partial or excessive action

2Measurement precision

If detailed recognition of all figures in the pattern is performed, then accurate shape determination is achieved, but computational complexity increases

Engineering Contradiction:
Improvefigure recognition accuracyVSAvoidcomputational complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The figure pattern is divided into multiple areas, each containing a subset of figures. The neural network processes each area separately and independently, breaking down the complex task of recognizing all figures in the entire pattern into smaller, parallelizable sub-tasks. This segmentation enables faster processing while maintaining accurate three-dimensional shape determination.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

Instead of requiring the neural network to recognize all figures in the complete figure pattern, the method uses only a subset of figures from selected areas. This partial action approach provides sufficient information for accurate three-dimensional shape reconstruction while significantly reducing computational complexity and processing time.

Inventive Principle:
Principle #16Partial or excessive action

Data Source

PatentEP3922942B1Method for determining a three-dimensional shape of at least one piece of a body and system for determining a three-dimensional shape of at least one piece of a body
Publication Date: 2023.03.29 OPTONIC GMBH
  • EP3922942B1 patent drawingFigure 1
  • EP3922942B1 patent drawingFigure 2
  • EP3922942B1 patent drawingFigure 3

AI summary

The invention relates to a method for determining a three-dimensional shape (FO) of at least one part (ST) of a body (KO), wherein the method comprises the steps: a) projecting a figure pattern (PM) onto at least the part (ST) of the body (KO), wherein the unprojected figure pattern (PM) has a region (BE) of recognizable figures (PU) repeating in one dimension (y), b) capturing an image (BI) of the projected figure pattern (PM) onto at least the part (ST) of the body (KO), c) determining figures (PU) in the captured image (BI) by means of recognition using a neural network (NN), wherein the neural network (NN) is trained by means of training images (TBI) of differently deformed sections (AS) of the figure pattern (PM) to recognize the figures (PU) in the region (BE), and by means of a region index (BEI) of the corresponding region (BE).and d) Determining the three-dimensional shape (FO) of at least the piece (ST) of the body (KO) based on the determined figures (PU).