Machine Vision Inspection Using Image Registration and Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Manual inspections of construction objects are time-consuming, costly, and prone to errors, and traditional image comparison methods struggle with distortions and irrelevant differences between inspection and reference images.
Innovation Solution
A system and method that uses image registration and neural networks, such as Faster Regional Convolutional Neural Networks (FRCNN), to compare inspection images with reference images, compensating for distortions and automatically identifying errors like part orientation and measurement errors, thereby enhancing inspection accuracy and efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manual inspection is performed by workers comparing objects to reference images, then inspection can be conducted with human judgment and correction capability, but the process becomes time-consuming, costly, and prone to errors
Solution Approach 1:
The patent replaces the manual mechanical inspection process with an automated machine vision system that captures images of the assembled object and reference object, processes them through computer algorithms, and automatically compares features to detect deviations. This substitution eliminates human time investment while maintaining or improving inspection accuracy through consistent algorithmic evaluation.
Solution Approach 2:
The patent creates digital image copies of both the assembled object and reference object, then performs comparisons on these copies rather than physical objects. This allows multiple analyses of the same inspection data without additional time cost, and enables automated processing that is faster and more consistent than manual review of the original objects.
2Productivity
If direct point-by-point comparison of inspection image and reference image is performed, then simple error detection is possible, but image distortions and viewing angle differences cause false positives and reduce reliability
Solution Approach 1:
The patent performs preliminary actions by detecting key features and establishing correspondence relationships between the inspection image and reference image before conducting the actual comparison. This includes identifying matching points, determining transformation parameters, and aligning the images in advance, which eliminates false positives from distortions while maintaining fast automated processing.
Solution Approach 2:
The patent changes parameters by applying geometric transformations (translation, rotation, scaling) to the inspection image based on detected feature correspondences. This adjusts the inspection image to match the reference image's coordinate system and viewing angle, enabling accurate comparison while maintaining automated efficiency.
3Productivity
If automated machine vision inspection is implemented, then inspection speed and consistency improve, but the system requires complex image processing algorithms and computational resources
Solution Approach 1:
The patent segments the inspection process into distinct modules: feature detection, correspondence establishment, image transformation, and deviation analysis. Each module handles a specific task with dedicated algorithms, which improves processing efficiency and makes the overall system more manageable despite the complexity of automated inspection.
Solution Approach 2:
The patent creates a universal image processing framework that can handle multiple types of objects and inspection requirements through configurable parameters and algorithms. The same core system performs feature detection, alignment, and comparison across different applications, reducing overall system complexity through reuse rather than requiring separate specialized systems for each inspection type.
Data Source
AI summary
An inspection method includes comparing an inspection image of an inspection object to a reference image of a reference object and recognizing at least one inspection part in the inspection image and at least one reference part in the reference image. The inspection part and the reference part correspond to each other, and the inspection image is registered onto the reference image using the inspection part and the reference part. A set of registration data is provided, and a checking for at least one error using the inspection image, the reference image, and the set of registration data is performed.