An improvement based on the hand pose model of YOLOv11-pose
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
Existing one-stage pose estimation models, such as the YOLO series, face challenges in balancing real-time performance and accuracy. Traditional convolutional operations suffer from spatial insensitivity and channel redundancy when extracting spatial features, limiting the model's ability to adaptively model the spatial context of keypoints. Meanwhile, conventional feature pyramid networks often employ simple addition or concatenation when fusing multi-scale features, ignoring the differences in contributions across different scales, leading to suboptimal information fusion. To address these issues, this paper modifies the official Hand Keypoints dataset from Ultralytics, obtaining 5000 images after data augmentation and processing. It proposes an innovative Involution-BiFPN collaborative augmentation network, YOLOv11n-BMT, which is seamlessly integrated into the latest generation YOLOv11-pose framework. The core idea of this design is to introduce an Involution operator in the critical stages of the backbone network to enhance the model's specific perception of the local spatial structure of pose keypoints; and to employ a weighted bidirectional feature pyramid network (BiFPN) in the neck network to achieve efficient and adaptive multi-scale feature fusion. The two work together to improve the model's accuracy in detecting human posture and locating key points in complex scenarios.
Fields of Study & Subjects
Human Pose and Action Recognition
Advanced Neural Network Applications
Hand Gesture Recognition Systems
Publication details
- Journal / source
- International Journal of Scientific Research and Management (IJSRM)
- Volume
- 14
- Issue
- 03
- Pages
- 2818–2827
- Publication date
- 1 January 2026
- Language
- EN