Inspection Target Segmentation for Camera Pose Planning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current automated visual quality inspection methods face challenges in accurately identifying inspection targets and determining optimal camera poses for effective visual inspection, particularly in manufacturing processes where defects can occur, leading to inconsistencies in quality assurance.
Innovation Solution
The method involves generating a spatial model of the item of manufacture using enrollment images from various camera poses, classifying regions to identify inspection targets, and calculating camera poses for improved inspection imaging, including the use of machine learning for target identification and geometric constraints to refine camera poses for enhanced inspection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If automated visual inspection methods are used to inspect manufactured items, then inspection efficiency is improved, but measurement precision of inspection targets deteriorates
Solution Approach 1:
The system performs preliminary classification of image regions to identify inspection targets before conducting detailed inspection. By pre-identifying and categorizing regions containing inspection targets using machine learning models, the system prepares data structures and selects appropriate inspection parameters in advance, ensuring both high processing speed and accurate target identification during the actual inspection process
Solution Approach 2:
The inspection process is divided into distinct stages: region classification to identify potential inspection targets, spatial model generation to locate targets precisely, and detailed inspection of identified targets. This segmentation allows each stage to be optimized independently - region classification uses efficient machine learning for speed, while spatial modeling ensures precise measurement accuracy for the final inspection
2Adaptability or versatility
If multiple camera poses are used to capture enrollment images from different angles, then inspection coverage is improved, but device complexity increases
Solution Approach 1:
The system transitions from managing multiple complex 3D camera poses to working with a unified 2D spatial model that represents all inspection targets in a standardized coordinate system. By projecting targets from various camera angles into a common 2D spatial representation, the system achieves comprehensive multi-angle inspection coverage while simplifying the mathematical complexity of pose management
Solution Approach 2:
The system creates a simplified 2D spatial model copy that represents the spatial relationships of inspection targets without requiring direct manipulation of complex 3D camera pose data. This spatial model copy serves as an intermediate representation that captures essential geometric information while being computationally simpler to work with than full 3D pose specifications
Data Source
AI summary
Automatic enrollment of an item of manufacture to a quality inspection system comprises using associations of enrollment images of an example of the item of manufacture to their corresponding camera poses. Enrollment images which show inspection targets (e.g., components of the item of manufacture) in views which are also suitable for use in visual inspection of further instances of the item of manufacture are identified. Their associated camera poses are selected and provided for use in inspection planning. In some embodiments, suitability of the camera pose is verified by performing inspection tests on the enrollment images.