Crawl Image Annotation for Real-World Structural Datasets
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing computer-vision-based structural health monitoring datasets are generated in controlled laboratory setups, making them unsuitable for real-life industrial applications, and manual annotation of images under varying conditions is time-consuming and costly.
Innovation Solution
A method and system for creating custom datasets by capturing images in a specified crawl direction, using pairwise quadratic transformations between sequential images, and modeling each crawl step with anchor points to automatically propagate bounding boxes across varying lighting conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manual annotation is used to create custom datasets, then dataset accuracy and reliability are improved, but time consumption and cost increase significantly
Solution Approach 1:
The system performs automatic annotation by using the crawl approach to generate bounding boxes and labels automatically. The method uses image matching and transformation models to propagate annotations across multiple images without human intervention, allowing the system to annotate itself rather than requiring manual annotation services.
Solution Approach 2:
The patent replaces the manual mechanical process of human annotation with an automated computational system. The system uses computer vision algorithms, image matching, and transformation models to automatically generate annotations, substituting the mechanical action of human annotators with automated digital processing.
2Measurement precision
If images are captured in controlled laboratory setups, then image quality and consistency are improved, but adaptability to real-life industrial applications deteriorates
Solution Approach 1:
The system dynamically adapts to varying capture conditions by using the crawl approach to capture images at multiple positions and orientations. The transformation models are updated based on actual capture data, allowing the system to adapt to real-life industrial variations in lighting, angle, and position while maintaining annotation accuracy.
Solution Approach 2:
The patent changes the parameters of image capture to include various crawl directions, positions, and orientations to represent real-life industrial conditions. By varying these capture parameters during data collection, the system creates a dataset that is adaptable to different real-world scenarios while maintaining sufficient image quality through the structured crawl approach.
3Ease of manufacture
If existing datasets are used for structural health monitoring, then model training is simplified, but the datasets are not suitable for real-life industrial applications
Solution Approach 1:
The crawl approach creates a universal dataset creation method that can be applied to various industrial scenarios. The system captures images at multiple crawl steps and generates annotations that are applicable across different real-life industrial conditions, making the dataset universally suitable for training models that need to perform reliably in diverse industrial environments.
Data Source
AI summary
A method of creating custom datasets is provided. The method comprises capturing, with a camera, a sequence of images of a surface in a specified crawl direction over a number of sequential crawl steps that represent changes of camera position. For each crawl step, the method matches a first image and a second image captured at sequential camera positions. The method models each crawl step as a pairwise quadratic transformation between the first image and second image according to a number of respective anchor points in each image selected as best candidates to fit a regression model.