Lighting Condition Setting for Machine Vision Inspection Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing methods for optimizing lighting conditions in machine vision systems for product appearance checking and facial recognition are time-consuming and require manual adjustments, often leading to suboptimal detection performance due to the need for manual regulation of lighting and check algorithms, and they do not effectively account for variations in workpieces or individuals.
Innovation Solution
A method that simultaneously optimizes lighting parameters and check algorithm parameters using a machine learning model, where objects are lit with variable lighting conditions, captured, and estimation images are generated and used to train the model, ensuring consistency with label data to improve detection accuracy and reduce system load.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual optimization of lighting and check algorithm is performed alternately and repeatedly, then detection performance can be improved, but time consumption increases significantly
Solution Approach 1:
The patent combines lighting parameter optimization and check algorithm optimization into a single unified process. The loss function simultaneously considers both lighting parameters (θL) and check algorithm parameters (θD), allowing both to be optimized together rather than alternately. This merging of optimization processes reduces the total time required while maintaining detection performance.
Solution Approach 2:
The patent introduces a lighting simulator as an intermediary component that generates estimation images based on lighting parameters. This simulator acts as a bridge between lighting configuration and image analysis, enabling gradient-based optimization of lighting parameters without requiring actual physical lighting changes and recaptures, thereby significantly reducing optimization time.
2Ease of operation
If lighting is regulated to be easily observed by the operator, then ease of operation improves, but checking accuracy may not be optimal
Solution Approach 1:
The system performs self-optimization of lighting parameters by automatically adjusting them to maximize detection accuracy through the unified loss function. The operator does not need to manually regulate lighting for ease of observation, as the system autonomously determines optimal lighting parameters (θL) that balance both observability and detection accuracy requirements.
3Adaptability or versatility
If optimization is performed on captured images of multiple workpieces to learn differences, then adaptability improves, but system calculation amount and time increase
Solution Approach 1:
The lighting simulator serves as an intermediary that enables efficient optimization by generating estimation images computationally rather than requiring physical recapture of multiple workpieces under different lighting conditions. This allows the system to learn from multiple workpieces and adapt to their differences while significantly reducing the time and computational resources required compared to physical experimentation.
4Adaptability or versatility
If there are many combined imaging and lighting parameters to optimize, then adaptability improves, but device complexity and optimization time increase
Solution Approach 1:
The patent replaces the mechanical/physical process of changing lighting conditions and recapturing images with a computational approach using a lighting simulator. The simulator computationally generates estimation images for different lighting parameters, substituting physical experimentation with mathematical modeling. This reduces device complexity and optimization time while maintaining the ability to explore many parameter combinations.
Data Source
Figure 1
Figure 2~3
Figure 4(a)~5
AI summary
The present disclosure relates to a method, device, system and program for setting a lighting condition when an object is checked and a storage medium. The method includes that: the object is lighted by light sources capable of changing lighting parameters specifying the lighting condition when the object is captured, and the object with corresponding label data is captured by an image sensor under multiple such lighting parameters to obtain multiple captured images; estimation images of the object are generated on the basis of image data sets obtained by associating the captured images and the corresponding lighting parameters; and the estimation images and the corresponding label data are applied to learning of the machine learning model, and the lighting condition is set on the basis of a comparison result between an estimation result of a machine learning model and the label data. Therefore, operations are simplified.