AR Image Training for Precise Object Segmentation and Contour Modeling

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing methods for detecting and modeling objects and characteristic areas in images suffer from precision issues due to the use of characteristic points, leading to inaccurate detection and modeling, and manual annotation lacks consistency and accuracy.

Innovation Solution

A machine learning system is trained using augmented reality images containing virtual elements to accurately segment and model objects and characteristic areas, employing a joint learning process that combines segmentation models with contour point sets, addressing occlusion problems and enhancing precision.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If traditional machine learning systems use separate processing stages for detection and recognition, then the system structure is simple and easy to implement, but the processing time increases and productivity decreases

Engineering Contradiction:
Improveprocessing speedVSAvoidsystem architecture complexity
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The patent combines multiple machine learning models (detection model and recognition model) into a single unified neural network architecture. This merging eliminates the need for separate processing stages, allowing the system to perform both detection and recognition simultaneously in one pass through the network, thereby reducing processing time while maintaining a manageable system structure through integrated design

Inventive Principle:
Principle #5Merging (Combining)

2Measurement precision

If the machine learning system processes entire high-resolution images, then the detection precision is high, but the computational resources and processing time increase significantly

Engineering Contradiction:
Improveobject detection precisionVSAvoidcomputational resource consumption
Core Design Contradiction:
Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The patent implements a two-stage processing approach where the image is first divided into multiple regions of interest based on initial detection, and then only these segmented regions are processed by the recognition model. This segmentation strategy maintains high detection precision by focusing computational resources on relevant areas while significantly reducing overall computational resource consumption compared to processing entire high-resolution images

Inventive Principle:
Principle #1Segmentation

3Reliability

If multiple machine learning models are trained separately on different datasets, then each model can be optimized for its specific task, but the training time and computational resources increase

Engineering Contradiction:
Improvemodel performanceVSAvoidtraining time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The patent merges multiple machine learning models into a single unified neural network that can be trained as one integrated system. This allows simultaneous optimization of detection and recognition capabilities during a single training process, reducing total training time while maintaining high model performance through joint learning and shared feature extraction layers

Inventive Principle:
Principle #5Merging (Combining)

Data Source

PatentEP4292062B1Learning method for a machine learning system for detecting and modelling an object in an image, corresponding computer program product and device
Publication Date: 2026.05.06 FITTINGBOX
  • EP4292062B1 patent drawingFigure 1~2d
  • EP4292062B1 patent drawingFigure 3~5
  • EP4292062B1 patent drawingFigure 6

AI summary

The invention relates to a learning method of a machine learning system. According to such a method, the machine learning system carries out the steps of: obtaining augmented reality images comprising a real image and at least one virtual element representative of the object and/or the characteristic region; obtaining, for each augmented reality image, learning information comprising, for at least one given virtual element of the augmented reality image: a model for segmenting the given virtual element, and a set of contour points corresponding to a parameterisation of the given virtual element; and learning on the basis of the plurality of augmented reality images and the learning information, delivering a set of parameters enabling the machine learning system to detect and model the object and/or the characteristic region in a given image.