3D Building Model Generation from 2D Images via Neural Network

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing methods for generating 3D digital models of building structures require extensive sensor data and manual measurements, making the process time-consuming and prone to errors due to human intervention.

Innovation Solution

A computer-implemented method using a 3D image generation neural network, specifically a dynamic resolution NeRF network, processes a series of 2D images or a video to extract key images, determine camera positions and directions, and generate metadata for a 3D digital model of a building structure.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If extensive sensor data and manual measurements are used to generate 3D digital models, then measurement precision and reliability are improved, but loss of time and device complexity increase

Engineering Contradiction:

Improveaccuracy of 3D digital modelVSAvoidtime-consuming process

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The patent replaces manual measurement processes and traditional sensor-based 3D scanning systems with an AI-powered image processing system. The neural network automatically extracts spatial information from standard 2D images, eliminating the need for mechanical measurement tools and manual data collection, thereby reducing time loss while maintaining measurement precision.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The system creates accurate 3D digital copies of building structures by processing standard 2D images. Instead of requiring specialized sensor data collection, the neural network generates precise 3D models by learning from and replicating spatial relationships present in ordinary images, significantly reducing the time and complexity of data acquisition.

Inventive Principle:

Principle #26Copying

2Measurement precision

If extensive sensor data and manual measurements are used to generate 3D digital models, then measurement precision is improved, but device complexity and ease of operation worsen

Engineering Contradiction:

Improveaccuracy of 3D digital modelVSAvoidcomplexity of data capture system

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The neural network system performs multiple functions using a single integrated approach: it detects spatial relationships, measures dimensions, generates 3D models, and extracts structural information all from standard 2D images. This universal system replaces multiple specialized devices and manual processes, reducing device complexity while maintaining measurement precision.

Inventive Principle:

Principle #6Universality (Multi-functionality)

Solution Approach 2:

The patent substitutes complex mechanical measurement systems and specialized sensor arrays with an AI-based image analysis system. The neural network processes standard images to extract precise spatial and dimensional information, eliminating the need for complex physical measurement equipment and simplifying the overall system architecture.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

3Reliability

If manual measurements and sensor data collection are performed, then reliability of data is improved, but ease of operation and productivity decrease

Engineering Contradiction:

Improvereliability of measurement dataVSAvoidspeed of 3D model generation

Core Design Contradiction:

ReliabilityVSProductivity

Solution Approach 1:

The neural network processes images continuously and automatically, extracting measurement data and generating 3D models without interruption. This continuous automated processing replaces discrete manual measurement steps, maintaining data reliability through consistent AI-based analysis while dramatically increasing productivity by eliminating human intervention delays.

Inventive Principle:

Principle #20Continuity of useful action

Solution Approach 2:

The system performs self-service by automatically detecting features, measuring dimensions, and generating 3D models from images without requiring manual measurement operations. The neural network independently extracts reliable spatial information and constructs accurate models, eliminating the need for human operators to perform time-consuming measurement tasks while maintaining data quality.

Inventive Principle:

Principle #25Self-service

Data Source

PatentUS20250148709A1Systems and methods in digital image processing for generating graphical three-dimensional models of the real-world environment from two or more two-dimensional images

Publication Date: 2025.05.08 HL ACQUISITION INC D B A HOSTA AI

AI summary

A computer-implemented method for generating a three-dimensional (3D) digital model from one or more two-dimensional images is described. The method includes: obtaining, though an application programming interface (API), a series of two-dimensional (2D) images of a scene taken by an image capturing device; extracting, by a processing device, key images from the series of 2D images, wherein each of the key images depicts one or more components of a building structure in the scene; determining, by the processing device, and based on the extracted key images, a respective position and a respective direction of the image capturing device relative to each of the one or more components of the building structure; and processing, using a 3D image generation neural network, the extracted key images and the positions and directions of the image capturing device to generate metadata comprising a three-dimensional (3D) digital model of the building structure.