Hand Posture Estimation via Key Point Feature Fusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing hand posture estimation technologies face challenges with inaccurate results due to unconstrained global and local posture variations, frequent occlusions, local self-similarity, and high intelligibility in hand images.
Innovation Solution
A hand posture estimation method that involves obtaining an initial feature map of a hand region, performing feature fusion processing to fuse features around key points, and then applying deconvolution processing to adjust the resolution of the fused feature map, ultimately determining the posture estimation result based on the coordinate information of the key points.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional hand posture estimation methods are used, then the processing speed is relatively fast, but the estimation accuracy is low due to unconstrained posture variations, occlusions, and local self-similarity
Solution Approach 1:
The patent segments the hand posture estimation problem by identifying and processing multiple key points (such as wrist, finger joints, and tip positions) independently. The feature fusion processing separately handles features around each key point, then integrates them to determine the overall posture. This segmentation approach improves accuracy by addressing local self-similarity and occlusion issues at each key point individually.
Solution Approach 2:
The patent merges features from multiple sources and multiple key points through feature fusion processing. It combines the initial feature map with features extracted around each key point, integrating local and global information. This merging approach resolves the contradiction by combining multiple feature representations to achieve higher estimation accuracy while managing complexity through systematic fusion.
Solution Approach 3:
The patent applies deconvolution processing to adjust and enhance the resolution of the fused feature map, effectively operating in a higher dimensional feature space. This dimensionality change allows the system to recover fine-grained spatial information that was lost during pooling operations, thereby improving key point localization accuracy without proportionally increasing processing complexity.
2Measurement precision
If feature fusion processing is applied to fuse features around multiple key points, then the posture estimation accuracy improves, but the computational complexity increases
Solution Approach 1:
The patent performs preliminary feature extraction and initial feature map generation before the main feature fusion processing. By preparing the initial feature representations in advance, the system reduces the computational burden during the fusion stage, as the complex operations have already been partially completed. This preliminary action allows accurate posture estimation while managing real-time computational power consumption.
Solution Approach 2:
The patent employs dynamic feature fusion that adapts to the specific characteristics of each key point and the current input image. Rather than applying uniform complex processing to all key points, the system dynamically adjusts the fusion operations based on local features and occlusion conditions, optimizing the balance between accuracy and computational power for each region.
3Measurement precision
If deconvolution processing is used to adjust the resolution of the fused feature map, then the localization precision of key points improves, but the processing time increases
Solution Approach 1:
The patent applies deconvolution processing selectively rather than uniformly across the entire feature map. It focuses the resolution enhancement primarily on regions containing key points, applying partial action where it is most needed for localization precision. This approach achieves high key point position precision while minimizing the overall processing time by avoiding excessive computation in non-critical regions.
Data Source
AI summary
Described are a hand posture estimation method, an electronic device, and a non-transitory computer-readable storage medium. The method includes: obtaining an initial feature map corresponding to a hand region in a candidate image; obtaining a fused feature map by performing feature fusion processing on the initial feature map; wherein the feature fusion processing is configured to fuse features around a plurality of key points; the plurality of key points represent skeleton key nodes of the hand region; obtaining a target feature map by performing deconvolution processing on the fused feature map; wherein the deconvolution processing is configured to adjust a resolution of the fused feature map; and obtaining coordinate information of the plurality of key points based on the target feature map to determine a posture estimation result of the hand region in the candidate image.