Neural Network Point Cloud Localization for Accurate 3D Alignment

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing localization methods in industrial metaverse environments, particularly for sensors like cameras, face challenges in accurately determining the 3D structure and pose of camera images, leading to inefficiencies in connecting the digital and real worlds.

Innovation Solution

A neural network-based apparatus and method that processes 3D point clouds using encoder and matching descriptor layers to calculate transformations between coordinate systems, enhancing point correlations and reducing deviations through training and fusion of feature vectors.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional point cloud processing methods are used for localization, then the system is simpler to implement, but the accuracy of transformation estimation deteriorates

Engineering Contradiction:
Improvetransformation estimation accuracyVSAvoidsystem complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent segments the point cloud processing into multiple hierarchical layers (coarse to fine), where each layer processes points at different levels of detail. This segmentation allows the system to achieve high transformation estimation accuracy by progressively refining matches across layers, while managing computational complexity through divide-and-conquer strategy.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent introduces intermediate feature descriptors and matching layers that act as mediators between raw point cloud data and final transformation estimation. These intermediate representations (encoded point clouds, feature vectors, matching scores) bridge the gap between input data and output transformation, improving accuracy while structuring the complex processing pipeline.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Measurement precision

If more point cloud features are processed to improve localization accuracy, then the measurement precision improves, but the computational time increases

Engineering Contradiction:
Improvelocalization accuracyVSAvoidprocessing time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent divides the point cloud into hierarchical layers with varying levels of detail. Coarse layers process fewer points quickly to establish initial matches, while fine layers process more points to refine accuracy. This segmentation reduces total processing time compared to uniformly processing all points at high detail, while maintaining localization accuracy through progressive refinement.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies partial processing strategies where not all point cloud features are processed to the same level of detail. Instead, the system processes sufficient features (excessive action relative to minimum needed) at appropriate hierarchical levels to achieve the required localization accuracy, avoiding unnecessary processing of redundant features that would increase time without improving results.

Inventive Principle:
Principle #16Partial or excessive action

3Measurement precision

If a single-layer processing approach is used, then the device complexity is reduced, but the localization accuracy in complex environments deteriorates

Engineering Contradiction:
Improvelocalization accuracyVSAvoidprocessing architecture complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent implements a multi-layer processing architecture where each layer handles specific aspects of point cloud analysis at different scales. This segmentation enables the system to capture both global structure (coarse layers) and local details (fine layers), achieving superior localization accuracy in complex environments compared to single-layer approaches, while organizing complexity into manageable modular layers.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent adds the hierarchical layer dimension to the processing architecture, transforming a single-level approach into a multi-level system. This dimensional expansion allows simultaneous processing of features at multiple scales, improving localization accuracy in complex environments by capturing both broad spatial relationships and fine-grained geometric details that single-layer systems miss.

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

Data Source

PatentEP4455611B1Localization by a neural network
Publication Date: 2026.02.25 NOKIA SOLUTIONS & NETWORKS OY
  • EP4455611B1 patent drawingFigure 1
  • EP4455611B1 patent drawingFigure 2
  • EP4455611B1 patent drawingFigure 3

AI summary

Method comprising: encoding a first 3D point cloud of a first coordinate system into a first encoded map comprising first feature center points and first feature vectors; encoding a second 3D point cloud of a second coordinate system into a second encoded map comprising second feature center points and second feature vectors; adapting the first input feature vectors based on the first input feature vectors and the second input feature vectors, to obtain a first joint map; adapting the second input feature vectors based on the first input feature vectors and the second input feature vectors to obtain a second joint map; calculating similarities between the first joint feature vectors and the second joint feature vectors and point correlations based on the similarities; checking whether a correlation condition is fulfilled; extracting the coordinates of the first joint feature center points and the coordinates of the second joint feature center points if the correlation condition is fulfilled; calculating a transformation between the first coordinate system and the second coordinate system based on the pairs of extracted coordinates.