Target Association via Attention Maps in Multi-View Image Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing methods for target association in computer vision require projecting targets into a common vector space, which is sensitive to image capture apparatus pose changes, leading to inaccuracies and errors in target correspondence determination.
Innovation Solution
An image processing method that determines attention maps for targets in multiple images and uses these maps to associate targets without requiring a common vector space, thereby reducing the impact of image capture apparatus pose changes and improving association accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If targets are projected into a common vector space for association, then target correspondence can be determined, but the system becomes sensitive to image capture apparatus pose changes, leading to inaccuracies
Solution Approach 1:
The patent introduces attention maps as an intermediary representation that bridges multiple images captured from different poses. Instead of directly projecting targets into a common vector space (which is sensitive to pose changes), the method uses attention maps to guide the association process, making the system more robust to pose variations while maintaining determination accuracy
Solution Approach 2:
The patent changes the parameter representation from fixed projection matrices (sensitive to pose) to dynamic attention maps that adapt to different image poses. By modifying how target positions are represented and compared across images, the system achieves both accuracy and robustness to pose changes
2Measurement precision
If projection matrix calibration is performed for target association, then correspondence determination can be achieved, but the process becomes complex and error-prone
Solution Approach 1:
The patent extracts the essential information needed for target association (attention maps indicating target positions and significance) from the images, eliminating the need for complex projection matrix calibration. By focusing only on the critical features for association rather than the complete geometric transformation, the method simplifies the process while maintaining accuracy
Solution Approach 2:
The patent uses attention maps that effectively copy and transfer target position information across different image representations without requiring explicit geometric calibration. This copying mechanism bypasses the complex projection matrix calibration process while preserving the necessary association information
Data Source
AI summary
The present disclosure relates to an image processing method and apparatus, an electronic device, and a storage medium. The method includes: acquiring at least two target images; determining an attention map of at least one target in each of the at least two target images according to a result of detecting target of each target image, where the attention map indicates the position of one target in a target image; and determining, based on each target image and the attention map of the at least one target in the each target image, a result of association that indicates whether a correspondence exists between at least some of targets in different target images.