Technical Drawing Graph Processing for Geometry-Annotation Association
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for processing technical drawings are inefficient in detecting and reconstructing technical data, such as geometry and annotations, from their images, limiting their use in generating accurate 3D models and manufacturing processes.
Innovation Solution
A computer-implemented method that constructs a graph representation of a technical drawing image, using machine-learning to identify and cluster nodes and edges, allowing for the reconstruction of geometries and associations between annotations and geometries, enabling the generation of a numerical model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional image processing methods are used to detect technical data from technical drawing images, then the processing can be performed with simple algorithms, but the detection precision and reconstruction accuracy of geometry and annotations are insufficient
Solution Approach 1:
The patent replaces traditional mechanical image processing algorithms with a neural network-based system. The neural network automatically learns features and patterns from technical drawing images, achieving superior detection precision for geometry and annotations without requiring manual feature engineering or complex processing pipelines.
Solution Approach 2:
The patent transforms the input image into a graph representation where pixels are nodes and spatial relationships are edges. This parameter transformation from raster image to graph structure enables the neural network to effectively capture both local and global spatial dependencies, improving reconstruction accuracy while maintaining computational efficiency.
2Productivity
If manual conversion of technical drawing images to numerical models is performed, then the accuracy can be maintained, but the productivity and efficiency are low
Solution Approach 1:
The patent implements an automated system where the neural network performs the entire conversion process from technical drawing image to numerical model without human intervention. The system automatically detects geometry, identifies annotations, reconstructs technical data, and generates the numerical model, achieving both high productivity and maintained accuracy through intelligent automation.
Solution Approach 2:
The patent introduces a graph representation as an intermediary between the input image and the final numerical model. This intermediate representation captures spatial relationships and structural information, enabling the neural network to accurately reconstruct geometry and annotations while automating the conversion process, thus maintaining precision while improving efficiency.
3Measurement precision
If detailed graph representation with all pixels as nodes is created, then the reconstruction accuracy can be improved, but the computational complexity and processing time increase significantly
Solution Approach 1:
The patent segments the image processing task by creating a graph where nodes represent meaningful spatial units rather than individual pixels. This segmentation reduces the number of nodes while preserving essential spatial relationships, enabling accurate reconstruction without the computational burden of processing every pixel as a separate node.
Solution Approach 2:
The patent extracts only the essential spatial relationship information from the image and represents it in the graph structure. By taking out and representing only the critical spatial dependencies rather than all pixel-level details, the system achieves good reconstruction accuracy with reduced computational complexity and faster processing time.
Data Source
Figure 1A
Figure 1A
Figure 1B
AI summary
The disclosure concerns a graph processing method. The graph represents an image of a technical drawing comprising a view and a technical annotation. The method comprises, for each view, providing the graph. The graph comprising nodes and edges. Each node corresponds to a classification of pixels into a semantic class of a predetermined set. Each edge connects two nodes either if the two nodes represent neighboring pixels or if the two nodes represent pixels distant from each other below a predetermined threshold. The set includes geometry, dimension, dimension-related symbol. The method comprises clustering, based on the graph topology: nodes corresponding to the geometry class, to reconstruct the geometries in the view, and nodes corresponding to the dimension and dimension-related symbol classes, to reconstruct the annotations of the view. The method comprises, associating reconstructed annotations to reconstructed geometries based on a detected position of the annotations and on the graph topology.