Virtual Pile Image Training for Accurate Workpiece Inference
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing techniques for generating inference models from images of piled workpieces are inefficient and lack accuracy in inferring workpiece information, particularly when dealing with varied viewpoints and complex pile configurations.
Innovation Solution
A system that generates virtual pile images from multiple workpiece images viewed from different angles, trains an inference model using machine learning on these images, and applies the model to real pile images to infer workpiece information, including segmentation, position, and orientation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing techniques are used to generate inference models from piled workpiece images, then the system can perform basic workpiece recognition, but the inference accuracy is insufficient and the process is inefficient when dealing with varied viewpoints and complex pile configurations
Solution Approach 1:
The patent creates virtual pile images by copying and synthesizing individual workpiece images into stacked configurations. This allows the generation of diverse training data showing various pile arrangements and viewpoints without requiring physical manipulation of actual workpieces, thereby improving inference accuracy while maintaining efficient model training
Solution Approach 2:
The system pre-generates virtual pile images with known ground truth information before training the inference model. By preparing these synthetic training datasets in advance with accurate labeling, the model can learn from diverse configurations efficiently, resolving the contradiction between accuracy and training efficiency
2Adaptability or versatility
If multiple workpiece images from different angles are processed to create virtual pile images, then the inference model can handle varied viewpoints, but the processing complexity increases
Solution Approach 1:
Instead of capturing multiple physical images from different angles, the system copies existing workpiece images and synthesizes virtual viewpoints through image processing. This approach maintains adaptability to various viewpoints while avoiding the complexity of multi-angle imaging hardware and synchronization systems
Solution Approach 2:
The patent transforms the problem from capturing images in physical space (multiple angles) to generating images in virtual space (synthetic compositions). By stacking 2D workpiece images to create virtual 3D pile representations, the system achieves viewpoint diversity without the complexity of multi-dimensional physical capture systems
Data Source
AI summary
A system includes circuitry configured to: generate a plurality of workpiece images each of which shows a workpiece viewed from a different viewpoint; generate, based on the plurality of workpiece images, one or more virtual pile images showing a plurality of piled workpieces; and generate, by machine learning using the one or more virtual pile images, an inference model configured to infer workpiece information regarding one or more of the workpieces shown in the pile image.