Optical Code Reading With ML Segmentation for Low-Contrast Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing code reading technologies struggle in dynamic scenarios due to variable object heights and changing contrasts, leading to reduced reading rates and increased false positives, especially when using conventional image processing and manual parameter settings.
Innovation Solution
A method utilizing a first segmentation process based on machine learning, particularly a neural network, to adapt code zone localization and decoding parameters dynamically, combined with classical image processing for enhanced accuracy and efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional image processing algorithms and manual classifiers are used for segmentation, then small code structures can be recognized, but a large number of false positive code zones are found in difficult reading situations
Solution Approach 1:
The system uses decoded code information as feedback to validate and filter code zone candidates. By attempting to decode detected code zones and using the results to refine detection parameters, the system reduces false positives while maintaining detection accuracy for small codes.
Solution Approach 2:
The segmentation parameters are made dynamic and adaptive rather than fixed. The system automatically adjusts segmentation thresholds and parameters based on the specific characteristics of each detected code zone, allowing it to distinguish between true codes and false positives in varying conditions.
2Reliability
If a high search contrast is required, then code zones can be reliably identified, but low-contrast code zones are overlooked
Solution Approach 1:
The search contrast threshold is made adaptive rather than fixed. The system dynamically adjusts the contrast threshold based on local image characteristics and code zone context, allowing it to reliably identify both high-contrast and low-contrast codes without manual parameter changes.
Solution Approach 2:
Different contrast thresholds and evaluation criteria are applied to different regions of the image based on local characteristics. Rather than using a global threshold, the system adapts its sensitivity to contrast variations in different parts of the image, improving both reliability and coverage.
3Adaptability or versatility
If a low search contrast is used, then low-contrast codes can be detected, but background textures are falsely considered as code zones
Solution Approach 1:
The system uses decoding feedback to validate low-contrast code zone candidates. By attempting to decode detected zones and using the results to refine detection, the system can confidently identify true low-contrast codes while filtering out background textures that fail to decode properly.
Solution Approach 2:
The system dynamically adjusts its confidence thresholds and validation criteria based on the detected code characteristics. For low-contrast detections, additional validation steps are automatically applied to distinguish true codes from background textures, reducing false positives.
4Device complexity
If segmentation parameters are fixed at the device side, then device complexity is reduced, but reading rate decreases due to unsuitable decoding settings
Solution Approach 1:
The system performs self-configuration and automatic parameter optimization without requiring manual intervention. By using machine learning models and automatic parameter tuning based on detected code characteristics, the system adapts to different applications while maintaining simple device operation.
Solution Approach 2:
Segmentation and decoding parameters are made dynamic and automatically adjusted based on the specific application scenario and detected code characteristics. The system transitions from fixed parameters to adaptive parameters that optimize reading rate for each situation without increasing user burden.
5Productivity
If the available decoding time is insufficient, then reading rate is indirectly reduced, but increasing decoding time for all code zones is not feasible
Solution Approach 1:
The system applies a two-stage approach where a fast initial segmentation identifies promising code zones, followed by more thorough decoding only for those candidates. This partial application of full decoding resources to all zones maintains high reading rate while ensuring accurate code recognition for true positives.
Solution Approach 2:
The system uses iterative feedback between segmentation and decoding stages to optimize time allocation. By using decoding results to refine segmentation and prioritize candidates, the system efficiently uses available decoding time to maximize the number of successfully read codes.
Data Source
AI summary
A method of reading optical codes is provided comprising the steps of recording an image, locating code zones in the image, and decoding the optical codes in the code zones, wherein the locating of code zones has a first segmentation process with machine learning by which first candidates for code zones are found. In this respect, the first candidates are evaluated to determine parameters for the locating of code zones and/or the decoding of the optical codes.