Machine Vision Training Images With Pre-Capture Annotation Feedback

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

The existing methods for providing annotated training images for machine vision systems are labor-intensive, time-consuming, and inefficient, particularly when retraining is required due to changes in image recognition tasks, and they often rely on large numbers of images that may not be useful for improving the algorithm's performance.

Innovation Solution

The method involves pre-annotating images before capture by the machine vision system, using the system to automatically generate annotated training images based on predefined annotation information, and applying the algorithm to test recognition accuracy, discarding images that meet the criteria, and iteratively refining the training process.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If manual annotation of training images is performed, then annotation accuracy is improved, but labor intensity and time consumption increase

Engineering Contradiction:
Improveannotation accuracyVSAvoidtime consumption
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The system performs preliminary action by automatically generating annotation information before the actual training image capture process. The annotation data is prepared in advance based on predefined annotation schemas, allowing the training images to be captured and annotated simultaneously without requiring subsequent manual annotation work, thus resolving the contradiction between annotation accuracy and time consumption

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The machine vision system performs self-service by automatically generating its own training data with annotation information without requiring external human annotators. The system uses its own capture device and processing capabilities to create annotated training images, eliminating the need for manual annotation while maintaining accuracy through automated annotation generation algorithms

Inventive Principle:
Principle #25Self-service

2Quantity of substance

If large numbers of images are collected for training, then training data quantity is improved, but storage requirements and processing complexity increase

Engineering Contradiction:
Improvetraining data quantityVSAvoidprocessing complexity
Core Design Contradiction:
Quantity of substanceVSDevice complexity

Solution Approach 1:

The system extracts only the essential annotation information needed for training from the captured images using predefined annotation schemas. Instead of processing and storing all image data extensively, the system extracts specific annotation features and properties that are sufficient for training the image content recognition algorithm, thereby reducing storage requirements and processing complexity while maintaining adequate training data quantity

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The system changes parameters by dynamically adjusting annotation schemas and capture settings based on the specific training needs. Rather than uniformly processing large numbers of images with fixed high-resolution requirements, the system adapts annotation depth, image resolution, and capture parameters to match the actual training requirements, reducing overall data volume and processing complexity while maintaining sufficient training data quantity

Inventive Principle:
Principle #35Parameter changes

3Measurement precision

If manual inspection and selection of training images is performed, then image quality is improved, but productivity decreases

Engineering Contradiction:
Improveimage qualityVSAvoidproductivity
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The system implements feedback by automatically evaluating captured images against predefined quality criteria and annotation schemas. The machine vision system provides real-time feedback on image quality and annotation completeness, enabling automatic acceptance or rejection of training images without manual inspection. This feedback mechanism maintains image quality standards while eliminating the time-consuming manual review process, thereby resolving the contradiction between image quality and productivity

Inventive Principle:
Principle #23Feedback

Solution Approach 2:

The system replaces the mechanical manual inspection process with automated computational evaluation. Instead of human operators visually examining and selecting training images, the machine vision system uses automated algorithms to assess image quality, verify annotation accuracy, and select appropriate training images. This substitution of mechanical human inspection with automated computational processes maintains quality control while dramatically improving productivity

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

4Adaptability or versatility

If retraining is performed due to task changes, then adaptability is improved, but time consumption and labor intensity increase

Engineering Contradiction:
ImproveadaptabilityVSAvoidretraining time
Core Design Contradiction:
Adaptability or versatilityVSLoss of time

Solution Approach 1:

The system performs preliminary action by maintaining a library of pre-generated annotation schemas and annotation templates that can be quickly applied when retraining is needed. When image recognition tasks change, the system can rapidly generate new training data by applying relevant pre-prepared annotation schemas without needing to create annotation frameworks from scratch, significantly reducing retraining time while maintaining adaptability to new tasks

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The system achieves universality by designing annotation schemas and data structures that can serve multiple image recognition tasks. The same annotation framework and data generation process can be applied across different recognition tasks by adjusting annotation parameters and schemas, allowing the system to efficiently retrain for new tasks using the same infrastructure, thereby reducing retraining time and labor intensity while maintaining adaptability

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

Data Source

PatentEP4654156A1Provision of annotated training images for use in training of a trainable image content recognition algorithm of a machine vision system
Publication Date: 2025.11.26 SICK IVP
  • EP4654156A1 patent drawingFigure 1A~1C
  • EP4654156A1 patent drawingFigure 2
  • EP4654156A1 patent drawingFigure 3A

AI summary

Method and arrangements for providing one or more annotated training images (336) for use in training of a trainable image content recognition algorithm of a machine vision system (205) that is operative, by means of the trainable image content recognition algorithm, to recognize content in images captured by the machine vision system (205). Annotation information (337) is obtained (301) for one or more intermediate images (338) to be captured by the machine vision system (205) and indicates properties of content in said intermediate images (338). The machine vision system (205) is operated (302) so that said one or more intermediate images (338) are captured by the machine vision system (205) in accordance with the obtained annotation information. It is provided (303) one or more annotated intermediate images (339) that correspond to said one or more intermediate images (338) captured by the machine vision system (205) with annotation in accordance with the obtained annotation information (337). Said one or more annotated training images (336) are then provided (306) based on at least one of said annotated intermediate images (339).