3D Map Inconsistency Detection via Multi-View Neural Networks

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Manual quality assurance of large-scale 3D maps is infeasible, leading to inconsistencies that degrade user experiences across various applications, and existing methods fail to enforce geometric consistency during 3D shape completion.

Innovation Solution

A 3D map inconsistency detection machine using a neural network that transforms 3D map data into 2D images compatible with standard CNNs, enabling robust and accurate detection of inconsistencies through an input transformation layer and ensemble neural networks.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If manual quality assurance is used for 3D maps, then detection accuracy can be high, but it becomes infeasible as the scale of 3D maps increases

Engineering Contradiction:
Improveinconsistency detection accuracyVSAvoidquality assurance efficiency
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The patent replaces manual quality assurance (mechanical human inspection) with an automated neural network system. The neural network processes 3D map data and multiple view images to detect inconsistencies automatically, eliminating the need for manual inspection while maintaining or improving detection accuracy. This substitution enables scalable quality assurance for large-scale 3D maps.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent introduces an intermediary transformation layer that converts 3D map data into multiple 2D view images from different virtual camera perspectives. This intermediary representation allows the neural network to analyze geometric consistency across different views, enabling automatic detection of inconsistencies without direct manual inspection of the complex 3D data.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Ease of manufacture

If existing 3D shape completion methods are used, then shape completion can be achieved, but geometric consistency among completed views is not enforced

Engineering Contradiction:
Improveshape completion capabilityVSAvoidgeometric consistency
Core Design Contradiction:
Ease of manufactureVSManufacturing precision

Solution Approach 1:

The patent implements a feedback mechanism where the neural network analyzes multiple views of the completed 3D shape and provides feedback on geometric consistency. The system processes images from different virtual camera views and uses the analysis results to detect and report inconsistencies, ensuring that the shape completion maintains geometric accuracy across all views.

Inventive Principle:
Principle #23Feedback

Solution Approach 2:

The patent transforms the 3D shape completion problem into a multi-dimensional analysis by generating and analyzing multiple 2D view images from different perspectives. This dimensional transformation allows the system to enforce geometric consistency by comparing the same 3D features across multiple 2D projections, detecting inconsistencies that would be difficult to identify in a single 3D representation.

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

3Stability of the object's composition

If 3D map data is directly processed, then the native 3D structure is preserved, but standard CNNs cannot be effectively applied

Engineering Contradiction:
Improve3D data structure integrityVSAvoidcompatibility with standard CNNs
Core Design Contradiction:
Stability of the object's compositionVSAdaptability or versatility

Solution Approach 1:

The patent applies dimensionality change by transforming 3D map data into multiple 2D view images that can be processed by standard CNNs. The transformation layer generates images from different virtual camera perspectives, allowing conventional 2D image processing networks to analyze 3D geometric consistency. This approach maintains compatibility with widely-used CNN architectures while preserving the essential 3D structural information through multi-view representation.

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

Data Source

PatentEP4162443B1Three-dimensional map inconsistency detection using a neural network
Publication Date: 2026.01.28 MICROSOFT TECHNOLOGY LICENSING LLC
  • EP4162443B1 patent drawingFigure 1~2
  • EP4162443B1 patent drawingFigure 3
  • EP4162443B1 patent drawingFigure 4

AI summary

A three-dimensional (3D) map inconsistency detection machine includes an input transformation layer connected to a neural network. The input transformation layer is configured to 1) receive a test 3D map including 3D map data modeling a physical entity, 2) transform the 3D map data into a set of 2D images collectively corresponding to volumes of view frustums of a plurality of virtual camera views of the physical entity modeled by the test 3D map, and 3) output the set of 2D images to the neural network. The neural network is configured to output an inconsistency value indicating a degree to which the test 3D map includes inconsistencies based on analysis of the set of 2D images collectively corresponding to the volumes of the view frustums of the plurality of virtual camera views.