Machine-Learning Image Inspection Triggers for Moving Objects
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image inspection systems struggle with setting flexible trigger conditions for moving objects, leading to inaccurate inspections due to missed triggers or unnecessary image processing, especially when objects deviate in position or angle within the capturing field of view.
Innovation Solution
An image inspection device that uses a machine learning model to set flexible trigger conditions by extracting feature amounts from frame images and determining inspection triggers based on a relative relationship between feature scores, allowing for robust and user-defined trigger settings.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a severe trigger condition is set to ensure only non-defective products trigger inspection, then inspection accuracy is improved, but inspection triggers are missed for defective products
Solution Approach 1:
The patent implements dynamic trigger conditions that adapt based on the inspection history and object characteristics. The system transitions from static, fixed trigger conditions to dynamic conditions that adjust sensitivity based on whether objects are detected or missed, allowing the trigger mechanism to become more permissive when objects are consistently missed and more selective when objects are reliably detected.
Solution Approach 2:
The system changes the trigger condition parameters dynamically based on inspection results. When an object is missed (no trigger generated), the system adjusts the trigger sensitivity parameter to be less severe. When objects are reliably detected, the parameter returns to a more selective state. This parameter adaptation resolves the contradiction between severe filtering and reliable detection.
2Adaptability or versatility
If a loose trigger condition is set to capture all potential inspection targets, then trigger application coverage is improved, but unnecessary inspections increase on images without objects
Solution Approach 1:
The system employs feedback from inspection results to adjust trigger conditions. When unnecessary inspections are performed (images without objects), the feedback mechanism reduces trigger sensitivity. When appropriate inspections are performed, the feedback maintains or increases sensitivity. This closed-loop feedback resolves the contradiction between comprehensive coverage and resource efficiency.
Solution Approach 2:
The trigger condition dynamically adjusts its looseness based on the detected pattern of unnecessary versus necessary inspections. The system transitions from a static, overly-permissive trigger condition to a dynamic condition that tightens or loosens based on real-time performance metrics, eliminating waste while maintaining adaptability.
3Loss of time
If external devices are used to provide trigger timing for image sensors, then inspection timing control is improved, but system complexity and setup time increase
Solution Approach 1:
The image processing device performs self-service by automatically generating inspection triggers based on its own image processing capabilities and inspection history. Instead of relying on external timing devices, the system uses its machine learning model's output and inspection results to autonomously determine when to trigger inspections, eliminating the need for complex external synchronization hardware and setup.
Solution Approach 2:
The system integrates multiple functions into a single device: image capture, machine learning inference, inspection result analysis, and trigger generation. This multi-functional integration eliminates the need for separate external trigger devices, reducing system complexity while maintaining precise timing control through the unified system's coordinated operation.
Data Source
AI summary
An image inspection device includes an image capturing section that generates a plurality of frame images aligned in time series, an inspection execution section that executes inspection processing of an object appearing in the plurality of frame images by a machine learning model to output an inspection result, and an inspection setting section that performs setting of the inspection execution section. The machine learning model includes a feature extraction section, and a determination section that outputs the inspection result from the feature amount. The inspection setting section receives selection of a first image, and determines a threshold. The inspection execution section outputs an inspection trigger when it is determined that the threshold is present between a first score based on a feature amount extracted from the first frame image and a second score based on a feature amount extracted from the second frame image.