Graphical Code Training Data for Alignment-Free Decoding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Decoding graphical codes such as barcodes requires proper alignment and positioning with dedicated scanners or cameras, necessitating operator expertise and complicating prompt decoding.
Innovation Solution
A method and system that generates learning data by creating graphical code images with various distortions and superpositions, allowing for accurate decoding without alignment or positioning, using a detection model trained on these varied images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If dedicated barcode scanners or cameras are used to decode graphical codes, then decoding accuracy is improved, but operator expertise and alignment/positioning requirements increase device complexity and operation difficulty
Solution Approach 1:
The system performs preliminary actions by pre-training the detection model with diverse distorted code images before actual decoding. This preliminary training enables the model to automatically adapt to various code presentations without requiring real-time alignment or positioning adjustments, thereby maintaining high decoding accuracy while eliminating complex operational requirements
Solution Approach 2:
The patent replaces the mechanical alignment and positioning system with an AI-based detection model. Instead of requiring physical adjustment of scanners or cameras to align with codes, the system uses machine learning to automatically detect and decode codes in various positions and orientations, substituting mechanical precision requirements with intelligent pattern recognition
2Measurement precision
If dedicated barcode scanners are used, then decoding accuracy is improved, but prompt and simple decoding deteriorates due to operator expertise requirements
Solution Approach 1:
The detection model performs self-service by automatically adapting to different code presentations through its pre-trained capabilities. The system serves itself by having the model internally handle various code distortions, rotations, and positions without requiring external operator intervention or expertise, thereby maintaining accuracy while simplifying operation
Solution Approach 2:
The patent substitutes the need for operator expertise with an automated detection model. The model replaces human judgment and manual adjustment skills with algorithmic pattern recognition, enabling any user to achieve accurate decoding results without requiring specialized training or experience
3Ease of manufacture
If standard code images are used for training, then training simplicity is improved, but detection model performance deteriorates due to lack of real-world distortion coverage
Solution Approach 1:
The patent applies segmentation by breaking down the training data generation process into distinct components: base code images, distortion transformations, and background elements. This segmented approach allows systematic creation of diverse training samples while maintaining control over each component, achieving both simplicity in generation and comprehensive coverage of real-world variations
Solution Approach 2:
The system employs parameter changes by systematically varying multiple parameters during training data generation, including rotation angles, scale factors, distortion types, background colors, and code positions. These parameter variations create a comprehensive dataset that covers diverse real-world scenarios, thereby improving model reliability while maintaining automated generation simplicity
Data Source
AI summary
A method of generating learning data includes: generating a first code image that is a graphical code; generating a second code image by applying first image processing to the first code image; and generating a first image by superposing the second code image onto a background image.