Disentangled Gait Recognition via Pose Feature Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing gait recognition methods using RGB cameras face challenges in capturing invariant gait features due to variations in clothing, viewing angles, and walking conditions, leading to sensitivity issues and loss or redundancy of gait information, especially with manual handcrafted features.
Innovation Solution
A novel approach is proposed to automatically disentangle dynamic pose features from pose-irrelevant features using an encoder-decoder network architecture, which extracts canonical and pose features from RGB video frames, allowing for robust gait recognition by comparing these features to stored sets for identification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Power
If appearance-based methods (e.g., GEI) are used for gait recognition, then computational cost is low and low-resolution imagery is handled well, but the method is sensitive to variations in clothing, carrying, and walking speed
Solution Approach 1:
The method segments gait features from appearance features by processing video frames to extract pose information that is independent of clothing and carrying variations. This segmentation allows the system to focus on motion patterns rather than appearance, resolving the contradiction between low computational cost and reliability under appearance variations.
Solution Approach 2:
The method extracts pose features (joint positions and movements) from video frames while discarding appearance information such as clothing color and texture. This extraction process removes the sensitivity to appearance variations while maintaining the essential gait information, achieving both low computational cost and high reliability.
2Reliability
If model-based methods using articulated body skeleton are used for gait recognition, then robustness to appearance variations is improved, but computational cost increases and dependency on pose estimation accuracy increases
Solution Approach 1:
The method uses a simplified pose estimation approach that focuses only on key body joints necessary for gait recognition, rather than full-body articulated modeling. This partial action reduces computational cost while maintaining robustness to appearance variations by capturing essential gait dynamics without the overhead of complex model-based methods.
3Loss of information
If manual handcrafted features are used for disentanglement, then gait features can be separated from appearance, but the method is sensitive to walking condition changes and may lose or create redundant gait information
Solution Approach 1:
The method uses dynamic pose estimation that adapts to different walking conditions by tracking joint positions and movements across multiple video frames. This dynamic approach automatically adjusts to variations in walking speed, carrying, and pose, maintaining effective disentanglement of gait features without the rigidity of manual handcrafted features.
Data Source
AI summary
Gait, the walking pattern of individuals, is one of the important biometrics modalities. Most of the existing gait recognition methods take silhouettes or articulated body models as gait features. These methods suffer from degraded recognition performance when handling confounding variables, such as clothing, carrying and viewing angle. To remedy this issue, this disclosure proposes to explicitly disentangle appearance, canonical and pose features from RGB imagery. A long short-term memory integrates pose features over time as a dynamic gait feature while canonical features are averaged as a static gait feature. Both of them are utilized as classification features.