2D Floorplan Visual Localization with Neural Ray Prediction

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing visual localization methods are complex, time-consuming, and computationally burdensome due to the construction of 3D representations of environments, which are costly in terms of storage and maintenance, and struggle with ambiguity in repetitive structures of floorplans.

Innovation Solution

Utilizing a 2D floorplan and a trained neural network to predict an array of rays from a camera's viewpoint, enabling efficient computation of 2D position and orientation without the need for 3D models, and employing a compact array of rays to search for consistent viewpoints.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If a 3D representation of the environment is constructed using point clouds from a depth sensor, then the camera localization accuracy is improved, but the computational burden and processing time increase significantly

Engineering Contradiction:
Improvecamera localization accuracyVSAvoidcomputational burden
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent extracts only the essential 2D floor plan representation from the environment, eliminating the need for complex 3D point cloud constructions. By using a simplified 2D representation with rays, the system achieves sufficient localization accuracy without the computational burden of full 3D modeling.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent transitions from 3D point cloud representations to a 2D floor plan representation. This dimensional reduction maintains the essential localization information while significantly reducing computational complexity and processing requirements.

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

2Measurement precision

If a 3D representation of the environment is constructed, then the camera localization accuracy is improved, but the storage requirements and maintenance costs increase

Engineering Contradiction:
Improvecamera localization accuracyVSAvoidstorage requirements
Core Design Contradiction:
Measurement precisionVSQuantity of substance

Solution Approach 1:

The patent extracts only the essential 2D floor plan representation from the environment, eliminating the need for complex 3D point cloud constructions. By using a simplified 2D representation with rays, the system achieves sufficient localization accuracy without the computational burden of full 3D modeling.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent transitions from 3D point cloud representations to a 2D floor plan representation. This dimensional reduction maintains the essential localization information while significantly reducing computational complexity and processing requirements.

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

3Device complexity

If a 2D floor plan with rays is used instead of a 3D representation, then the processing load and bandwidth requirements are reduced, but the ability to handle complex environments may be compromised

Engineering Contradiction:
Improveprocessing loadVSAvoidenvironmental adaptability
Core Design Contradiction:
Device complexityVSAdaptability or versatility

Solution Approach 1:

The patent creates a universal 2D floor plan representation that can be applied across different environment types. The ray-based approach provides a generalizable framework that handles various indoor spaces without requiring environment-specific 3D models, thereby maintaining versatility while reducing processing complexity.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Data Source

PatentEP4592956A1Visual localization
Publication Date: 2025.07.30 MICROSOFT TECHNOLOGY LICENSING LLC
  • EP4592956A1 patent drawingFigure 1
  • EP4592956A1 patent drawingFigure 2
  • EP4592956A1 patent drawingFigure 3

AI summary

Examples describe using a two dimensional floor plan, rather than a three dimensional (3D) reconstruction of a scene, to localize a camera. A floorplan of an environment is accessed and an image of the environment captured by a camera in the environment is received. The image is input to a trained neural network to predict an array of rays from the camera to surfaces in the environment indicated in the floorplan. Using the array of rays it is then possible to compute 3D position and orientation of the camera with respect to the floorplan.