3D Model Indexing via 2D Image-Voxel Correlation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current methods for generating 3D models from 2D images lack efficient association of video data with areas of interest, requiring manual annotation and navigation by experts, which is time-consuming and inefficient, especially in applications like inspection of real-world objects in extreme environments or complex industrial settings.

Innovation Solution

A system that indexes 2D images and 3D models by correlating pixels with voxels, allowing for direct annotation and association of video data with specific features in the 3D model, enabling automatic retrieval of relevant image data and reducing the burden on experts by providing bidirectional correspondence between 2D images and 3D models.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If manual annotation and navigation of video footage is used to identify areas of interest, then experts can detect issues with objects, but the process is time-consuming and inefficient

Engineering Contradiction:
Improveissue detection accuracyVSAvoidinspection time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The system pre-processes video footage to automatically generate 3D models with annotated areas of interest before expert review. Operators annotate videos during capture, and the system automatically processes these annotations into structured 3D model data, preparing the inspection results in advance for efficient expert analysis.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent introduces an intermediary processing system that automatically converts video annotations into 3D model annotations. This intermediary layer translates operator notes and video frame references into spatially-accurate 3D model annotations, eliminating the need for experts to manually navigate video footage while preserving detection accuracy.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Reliability

If experts manually review hours of video footage to detect issues, then comprehensive inspection is possible, but the workload and time required increase significantly

Engineering Contradiction:
Improveinspection completenessVSAvoidinspection efficiency
Core Design Contradiction:
ReliabilityVSProductivity

Solution Approach 1:

The system segments the inspection process into distinct phases: video capture with operator annotation, automatic 3D model generation, and expert review of annotated 3D models. This segmentation allows each phase to be optimized independently, with automation handling time-consuming processing while experts focus on analysis.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent creates a 3D model copy of the inspected object that preserves all spatial and contextual information from the video footage. Experts can review this 3D representation instead of original video, maintaining inspection completeness while dramatically reducing review time and cognitive load.

Inventive Principle:
Principle #26Copying

3Productivity

If 3D models are generated from 2D video images, then visual data is processed efficiently, but automatic association of video data with areas of interest is difficult without manual annotation

Engineering Contradiction:
Improvedata processing efficiencyVSAvoidannotation system complexity
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The system implements feedback loops where operator annotations on video footage are automatically processed and reflected in the 3D model annotations. The system provides feedback by displaying annotated 3D models to operators for verification, creating an iterative refinement process that improves accuracy without requiring complex manual intervention.

Inventive Principle:
Principle #23Feedback

Solution Approach 2:

The patent enables the system to automatically perform annotation transfer from video to 3D models without requiring expert intervention. The automated processing system serves itself by converting video frame annotations into 3D model annotations using the established correspondence data, reducing the need for manual annotation efforts.

Inventive Principle:
Principle #25Self-service

4Loss of information

If extensive video footage is stored for inspection review, then all inspection data is preserved, but retrieving specific areas of interest requires manual searching through large datasets

Engineering Contradiction:
Improveinspection data preservationVSAvoiddata retrieval time
Core Design Contradiction:
Loss of informationVSLoss of time

Solution Approach 1:

The system transforms 2D video data into a 3D model representation, adding spatial dimensionality to the data structure. This dimensional transformation organizes inspection data in a more intuitive and searchable format, where areas of interest can be directly located in 3D space rather than searched through linear video timelines.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

Solution Approach 2:

The system pre-establishes correspondence data between video frames and 3D model voxels during the 3D model generation process. This preliminary indexing allows rapid retrieval of specific areas by spatial coordinates or annotation references, eliminating the need for manual video searching while preserving all original inspection data.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP3161412B1Indexing method and system
Publication Date: 2021.07.14 WHITECAP SCI
  • EP3161412B1 patent drawingFigure 1
  • EP3161412B1 patent drawingFigure 2a~2c
  • EP3161412B1 patent drawingFigure 3a~3b

AI summary

A method is disclosed for capturing 3D model data including data relating to each of a plurality of voxels and relating to an object. A plurality of images of the object are captured. The plurality of images are correlated with the 3D model data to produce index data, the index data for indicating a correlation between some of the plurality of images and some of the plurality of voxels. The index data is then stored.