Feature Map Grouping for Multi-Person Pose Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing systems face increased processing load and reduced accuracy in posture estimation when multiple individuals are present in an image, as they require generating feature maps for each person, limiting their applicability.
Innovation Solution
Generate a first feature map specifying the horizontal location and a second feature map specifying the vertical location of individuals in an image, using these maps to associate detected joints with corresponding persons through a grouping process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If feature maps are generated for each person in the image, then joint-person association accuracy is improved, but processing load increases with the number of people
Solution Approach 1:
The patent segments the feature map generation process by creating separate horizontal and vertical component maps instead of generating complete feature maps for each person. This segmentation allows the system to process location information in distributed components that can be combined later, reducing the computational burden of generating full feature maps for multiple individuals simultaneously.
Solution Approach 2:
The patent transforms the traditional single-dimension feature map approach into a two-dimensional structure by separating horizontal and vertical location information into distinct component maps. This dimensional change enables more efficient processing and storage of location data, as the system only needs to maintain and process two component maps regardless of the number of people in the image.
2Measurement precision
If feature maps are generated for each person in the image, then joint detection accuracy is improved, but device complexity increases
Solution Approach 1:
The patent segments the complex feature map generation task into simpler horizontal and vertical component maps. Instead of implementing a complex system that generates complete feature maps for each individual, the system uses segmented component maps that capture essential location information with reduced computational and structural complexity.
Solution Approach 2:
The horizontal and vertical component maps serve as universal structures that can be used for all persons in the image simultaneously. These component maps function as multi-purpose data structures that provide location information for any number of individuals without requiring separate processing paths or additional system components for each person.
Data Source
AI summary
An image processing apparatus includes a feature map generation unit and a grouping unit. The feature map generation unit generate a first feature map that specifies a horizontal location of the person in an image in a horizontal direction and a second feature map that specifies a vertical location of the person in the image in a vertical direction. The grouping unit uses locations of joints detected in the image in the horizontal direction and vertical direction, the first feature map, and the second feature map, to group each of the joints to a corresponding person.