Physical Markers for Automated Image Labeling of Local Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for generating labeled training data for machine learning models, particularly for classifying local features of objects, are time-consuming, costly, and inefficient, especially for objects with reflective or transparent surfaces and varying lighting conditions, requiring large sets of manually labeled data.
Innovation Solution
A method using physical marker devices, such as AR or QR markers, is applied adjacent to the local features to facilitate the acquisition of images, allowing for automated generation of training data by detecting and computing regions of interest, which are then used to train models for classification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual labeling by human experts is used to generate training data, then labeling accuracy can be maintained, but the process becomes time-consuming and costly
Solution Approach 1:
Physical marker devices are introduced as intermediaries between the object and the labeling process. These markers serve as detectable references that guide automated systems in identifying and labeling regions of interest, replacing the need for manual expert labeling while maintaining accuracy
Solution Approach 2:
The manual mechanical process of human experts visually inspecting and drawing labels is replaced by an automated optical system using camera sensors to detect physical markers and compute regions of interest programmatically
2Reliability
If large sets of training data are generated to improve model reliability, then classification accuracy improves, but the cost and time investment increases significantly
Solution Approach 1:
Instead of manually creating diverse training samples, the system uses physical markers that can be rapidly applied to multiple objects and configurations. Camera sensors capture images of these marked objects, automatically generating large volumes of training data through efficient copying and variation of marked samples
Solution Approach 2:
The system enables rapid generation of training data by varying parameters such as object positions, lighting conditions, and marker placements. This allows efficient creation of diverse training datasets that improve model reliability without proportionally increasing time and cost investment
3Productivity
If simulation and rendering are used to generate training data, then large amounts of labeled data can be produced without manual labeling, but the data differs from real world conditions and requires complex tools
Solution Approach 1:
Physical marker devices serve as intermediaries that bridge the gap between real-world objects and automated detection. These tangible markers provide reliable detection references in real-world conditions without requiring complex simulation environments, enabling direct capture of realistic training data
Solution Approach 2:
Complex simulation and rendering software systems are replaced by a simpler physical approach using tangible marker devices and camera sensors. This substitution maintains data realism while achieving high productivity through direct physical capture rather than virtual generation
Data Source
AI summary
The disclosure concerns generating image data for generating training information, and generating the training information for an automated image analysis related to a local feature in an image. A method comprises applying at least one physical marker device adjacent to the local feature of an object, acquiring a plurality of images of the object, and detecting the at least one physical marker device in at least one image. For each detected physical marker device, a region of interest in the at least one image based on predetermined relative location information associated with the at least one physical marker device is computed, mask information based on the computed region of interest is generated and stored associated with the at least one image as the training information. Classification information for detecting, segmenting, classifying, identifying, or determining a regression for the local feature is generated by training a model using the training information.