Occluded Pedestrian Re-Identification via Pose-Guided Feature Alignment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing occluded pedestrian re-identification methods face challenges in robust feature expression due to occlusions, where the introduction of occlusion features reduces recognition ability and misalignment of local features occurs, and attention mechanisms are insufficient in addressing background clutter.
Innovation Solution
The method employs a graph convolutional module for local feature enhancement and an attention-guided background suppression module, using pose estimation and heat maps to focus on visible pedestrian parts, integrating context information and reducing background influence.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If occlusion features are introduced into the model, then the model can handle occluded pedestrians, but the recognition ability is reduced
Solution Approach 1:
The patent divides the pedestrian image into multiple local regions based on pose estimation keypoints (head, upper body, lower body, etc.). Each local region is processed separately to extract local features, avoiding the negative impact of occlusion on the entire image while maintaining recognition ability through selective regional analysis.
Solution Approach 2:
The patent applies different processing strategies to different local regions based on their occlusion status. Visible regions are weighted more heavily, while occluded regions are suppressed or excluded from feature extraction. This local quality differentiation ensures that reliable features drive recognition while occlusion features do not degrade performance.
2Reliability
If local pedestrian features are matched, then the recognition ability is improved, but occlusion leads to misalignment of local features
Solution Approach 1:
The patent performs pose estimation and keypoint detection before local feature extraction and matching. By pre-identifying the spatial locations of body parts through pose estimation, the system establishes correct correspondence between local features of different pedestrians, ensuring accurate alignment even in occluded conditions where direct feature matching would fail.
3Object-affected harmful factors
If attention mechanism is used to weight visible parts, then background clutter impact is reduced, but model complexity increases
Solution Approach 1:
The patent segments the pedestrian image into multiple local regions based on semantic body parts (head, upper body, lower body, etc.). This segmentation allows the attention mechanism to operate on discrete, meaningful regions rather than the entire image, reducing the computational burden while effectively suppressing background clutter in each region.
Data Source
AI summary
The present application relates to an occluded pedestrian re-identification method, including steps of obtaining global features and local features of occluded pedestrians, and recombining the local features into a local feature map; obtaining a heat map of key-points of pedestrian images and a group of key-point confidences, obtaining a group of features of the pedestrian key-points by using the local feature map and the heat map; obtaining a local feature group by using the global features to enhance each key-point feature in the group of features of pedestrian key-points according to Conv, and an adjacency matrix of key-points is obtained through the key-points, the local feature group and the adjacency matrix of key-points are used as the input of GCN to obtain the final features of pedestrian key-points.