Contextual Instance Decoupling for Robust Multi-Person Pose Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing multi-person pose estimation (MPPE) methods face challenges such as wrong bounding box clipping, wrong assembly of keypoints, and long-distance regression, making them non-robust and non-real-time, which are critical for intelligent human information collection and perception in smart city and digital retinal architectures.
Innovation Solution
A CID-based MPPE model with an instance information abstraction module, global feature decoupling module, and heatmap estimation module is used to estimate pose in images, decoupling instance features and generating probability distributions of keypoints, enhancing robustness to spatial detection errors and alleviating keypoint grouping difficulties.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If top-down estimation approach is used, then pose estimation can be performed, but wrong bounding box clipping occurs reducing accuracy
Solution Approach 1:
The patent segments the pose estimation process into instance information abstraction, global feature decoupling, and heatmap estimation modules. Each module handles specific aspects independently, allowing the system to process multiple persons without wrong bounding box clipping while maintaining accuracy through dedicated instance features for each person.
Solution Approach 2:
The patent introduces instance feature perception maps as intermediary representations between the input image and final pose estimation. These perception maps serve as mediators that correctly associate keypoints with their corresponding persons, eliminating wrong assembly issues and improving robustness to spatial detection errors.
2Measurement precision
If bottom-up estimation approach is used, then pose estimation can be performed, but wrong assembly of keypoints occurs reducing accuracy
Solution Approach 1:
The patent introduces instance feature perception maps as intermediary representations between the input image and final pose estimation. These perception maps serve as mediators that correctly associate keypoints with their corresponding persons, eliminating wrong assembly issues and improving robustness to spatial detection errors.
Solution Approach 2:
The patent applies local quality by generating dedicated instance features for each person through spatial and channel recalibration. Each person's keypoints are processed with instance-specific features, ensuring correct assembly even in crowded scenes where global features would cause confusion.
3Productivity
If single-stage regression approach is used, then pose estimation can be performed, but long-distance regression reduces reliability
Solution Approach 1:
The patent segments the feature processing into instance-specific feature extraction and decoupling stages. By creating separate instance feature perception maps for each person, the system performs regression on localized instance features rather than long-distance global features, improving reliability while maintaining real-time speed through efficient modular processing.
4Device complexity
If global feature map is used for all persons, then processing is simplified, but keypoint grouping difficulty increases reducing accuracy
Solution Approach 1:
The patent segments the global feature map into instance-specific feature perception maps through instance information abstraction and global feature decoupling modules. This segmentation maintains relatively simple model structure while significantly improving keypoint assembly accuracy by providing dedicated features for each person.
Solution Approach 2:
The patent applies local quality by generating dedicated instance features for each person through spatial and channel recalibration. Each person's keypoints are processed with instance-specific features, ensuring correct assembly even in crowded scenes where global features would cause confusion.
Data Source
AI summary
The present application relates to the technical field of deep learning and pose estimation, and more particularly, to a contextual instance decoupling (CID)-based multi-person pose estimation (MPPE) method and apparatus. The method includes: acquiring a preset number of images containing multiple persons; inputting the images containing multiple persons, as a training sample, into a CID-based MPPE model for training; and performing pose estimation on a target image using the trained CID-based MPPE model, the CID-based MPPE model being provided with an instance information abstraction module, a global feature decoupling module and a heatmap estimation module. The method and apparatus of the present application can explore context clues over a greater range, thus being robust to spatial detection errors and superior in both accuracy and efficiency.