2D Image Processing for 3D Model Generation and 2D Floor Plan Creation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional methods for generating 2D floor plans require manual user input and may be cumbersome, leading to inaccuracies, especially when using non-LiDAR technology like smartphones.

Innovation Solution

The use of a computer system that employs 2D image processing to generate a 3D model of a room, which is then corrected and projected onto a 2D plane to create an accurate floor plan, leveraging machine learning models for semantic segmentation and trajectory-based room delineation.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If manual user input methods are used to generate 2D floor plans, then the process can be completed with simple technology, but the process becomes cumbersome and time-consuming

Engineering Contradiction:
Improvefloor plan generation speedVSAvoiduser input time
Core Design Contradiction:
ProductivityVSLoss of time

Solution Approach 1:

The system automatically captures images, generates 3D models, extracts 2D floor plans, and detects furniture objects without requiring manual user input for each step. The processing system serves itself by autonomously completing the entire floor plan generation workflow from raw images to final annotated plans.

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The system performs preliminary 3D model generation and 2D projection before furniture detection. By pre-processing the images to create 3D models and initial 2D floor plans, the system prepares the data structure in advance, making the subsequent furniture detection more efficient and accurate.

Inventive Principle:
Principle #10Preliminary action

2Device complexity

If non-LiDAR technology like smartphones is used, then the device complexity is reduced, but the measurement precision deteriorates

Engineering Contradiction:
Improvedevice hardware requirementsVSAvoidfloor plan accuracy
Core Design Contradiction:
Device complexityVSMeasurement precision

Solution Approach 1:

The system replaces LiDAR-based mechanical 3D scanning with 2D image-based computational 3D reconstruction. Instead of using light detection and ranging hardware, the system uses multiple 2D images captured by standard smartphone cameras and reconstructs 3D models through image processing and machine learning algorithms.

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

Solution Approach 2:

The system transforms 2D image parameters into 3D spatial parameters through computational methods. By changing the parameter representation from 2D coordinates to 3D model data, the system achieves accurate measurements using standard camera hardware that would otherwise only capture 2D information.

Inventive Principle:
Principle #35Parameter changes

3Ease of manufacture

If conventional 2D plan generation methods are used, then the process is simple, but the accuracy of the floor plan deteriorates

Engineering Contradiction:
Improveprocess simplicityVSAvoidfloor plan accuracy
Core Design Contradiction:
Ease of manufactureVSManufacturing precision

Solution Approach 1:

The system introduces a 3D intermediate dimension to improve 2D floor plan accuracy. By generating 3D models from 2D images and then projecting back to 2D, the system uses the additional spatial dimension to capture depth information and geometric relationships that would be lost in direct 2D processing, thereby improving measurement accuracy.

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

Data Source

PatentUS20250200873A1Two dimensional image processing to generate a three dimensional model and determine a two dimensional plan
Publication Date: 2025.06.19 AMAZON TECH INC
  • US20250200873A1 patent drawing
  • US20250200873A1 patent drawing
  • US20250200873A1 patent drawing

AI summary

Techniques for two-dimensional (2D) image processing to generate a three-dimensional (3D) model and determine a 2D plan are described herein. In an example, a 3D model of a room can be generated by using a video file portion of a video file as a first input to a first machine learning (ML) model. Semantic segmentation of the room can be generated by using the video file portion as a second input to a second ML model. The semantic segmentation may indicate that an object having an object type is shown in a first image frame of the video file portion. A 3D representation of the object in the 3D model can be determined. The 3D model can be corrected by setting a property of the 3D representation to a predefined value. A 2D floor plan of the room can be generated based on the corrected 3D model.