Multi-CNN Feature Extraction for Accurate Image Projection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing environment recognition systems using a single convolutional neural network (CNN) for calculating correspondence relationships between images suffer from low accuracy in determining the correspondence relationship between images.
Innovation Solution
A method involving multiple CNNs for extracting feature points from different images and deriving correspondence relationships, followed by geometric correction and projection using a control device to enhance accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single CNN is used for feature extraction and correspondence relationship determination, then the device complexity is reduced, but the measurement precision of correspondence relationship is low
Solution Approach 1:
The patent divides the correspondence relationship determination into separate functional modules: a first CNN for extracting feature points from the material image, a second CNN for extracting feature points from the object image, and a third CNN for determining correspondence relationships between the extracted feature points. This segmentation allows each network to specialize in specific tasks, improving overall measurement precision while maintaining manageable complexity through functional decomposition.
Solution Approach 2:
The patent introduces an intermediary feature point extraction stage using multiple CNNs before the final correspondence relationship determination. The first and second CNNs act as intermediaries that process the material and object images separately to extract feature points, which are then fed into the third CNN for correspondence analysis. This intermediary processing layer improves the quality of input data for the final matching operation.
2Measurement precision
If multiple CNNs are used for feature extraction and correspondence relationship determination, then the measurement precision is improved, but the device complexity increases
Solution Approach 1:
The patent segments the vision processing system into three distinct CNN modules, each responsible for a specific function: material image feature extraction, object image feature extraction, and correspondence relationship determination. This segmentation enables specialized processing for each task, improving measurement precision while organizing complexity into modular, manageable components.
Solution Approach 2:
The patent creates a multi-functional processing pipeline where the extracted feature points from the first and second CNNs serve as input to the third CNN for correspondence analysis. This universal approach allows the system to handle multiple processing stages within a unified architecture, improving precision without proportionally increasing complexity.
3Manufacturing precision
If feature points are extracted using multiple CNNs and geometric correction is performed, then the manufacturing precision of projection image is improved, but the calculation time increases
Solution Approach 1:
The patent segments the image processing pipeline into distinct stages: material image feature extraction using a first CNN, object image feature extraction using a second CNN, correspondence relationship determination using a third CNN, and geometric correction. This segmentation allows for optimized processing at each stage and enables parallel computation where appropriate, improving manufacturing precision while managing calculation time through structured processing.
Solution Approach 2:
The patent performs preliminary feature point extraction using the first and second CNNs before the correspondence relationship determination. By pre-processing the material and object images to extract feature points in advance, the system prepares the necessary data for the final matching operation, improving the precision of the projection image while optimizing the overall calculation sequence to minimize time loss.
Data Source
AI summary
A calculation method includes extracting a plurality of first feature points by performing a first convolution operation on a material image which is input to a first CNN, and which is to be projected onto an object, extracting a plurality of second feature points by performing a second convolution operation on an object image which is input to a second CNN, and which includes the object, and deriving a correspondence relationship between a plurality of first corresponding points and a plurality of second corresponding points by performing a third convolution operation on the plurality of first corresponding points belonging to the plurality of first feature points input to a third CNN, and on the plurality of second corresponding points belonging to the plurality of second feature points input to the third CNN.