MA-Pose and DMMS-GCN: skeleton-based martial-arts category recognition with edge-device deployment
- 1Department of Physical Education, Suzhou Vocational Health College, Suzhou, Jiangsu, China
- 2Department of Physical Education, Tangshan Normal University, Tangshan, China
- 3Department of Physical Education, Woosuk University, Wanju-gun, Jeollabuk-do, Republic of Korea
Abstract
With the development of visual sensor technology and human pose estimation algorithms, skeleton-based action recognition has received extensive attention in sports analysis and training systems. The recognition of martial-arts actions is difficult because different styles can share similar poses while differing in rhythm and coordination. To support this task, we constructed the Martial Arts Pose (MA-Pose) dataset, which contains 433 source videos across 13 martial arts categories. The 33-landmark sequences include normalized image coordinates and monocular relative depth rather than metric 3D ground truth. We further propose a Dynamic Multi-Stream Multi-Scale Graph Convolutional Network (DMMS-GCN), which integrates joint, bone, and motion features. The model combines physical, learnable, and input-adaptive graph topology with parallel temporal convolutions. With a fixed seed of 42, 346 videos are used for training and 87 for validation, with one center crop and one prediction per validation video. DMMS-GCN correctly classifies 73/87 videos (83.91% Top-1; 83.51% Macro-F1; 95% Wilson confidence interval, 74.78%–90.17%) and retrieves the correct class among the top five for 84/87 videos (96.55%). These values describe development-set performance because model selection and evaluation use the same validation subset. On a Raspberry Pi 4B, pose extraction runs at approximately 10 frames/s and classification of a prepared 150-frame tensor takes approximately 120 ms. Collecting the first 150 processed pose frames takes about 15 s, so the deployment demonstrates edge-device feasibility rather than instantaneous recognition. The source code, model checkpoint, and MA-Pose dataset are publicly available.
Related
- CNN-based dual-morphology mosquito recognition and mobile deployment
- Object recognition in autism: Shape configuration and Category-specificity
- Fine-grained Chinese dish recognition on CNFOOD-241: A discriminative learning approach for improving category-level discrimination
- The Influence of Mindfulness Meditation on Anxiety Levels Among Martial Arts Athletes in Facing Competitions
- Effect of soft Chinese martial arts on depressive symptoms among college students: a systematic review and meta-analysis
BibTeX
@article{wang2026mapose,
title = {{MA-Pose and DMMS-GCN: skeleton-based martial-arts category recognition with edge-device deployment}},
author = {Longying Wang and Ruijie Zhang and Huiteng Xu and Shilin Guo},
journal = {Frontiers in Neurorobotics},
year = {2026},
month = nov,
doi = {10.3389/fnbot.2026.1967140},
url = {https://doi.org/10.3389/fnbot.2026.1967140},
note = {CC BY},
}
RIS
TY - JOUR
TI - MA-Pose and DMMS-GCN: skeleton-based martial-arts category recognition with edge-device deployment
AU - Longying Wang
AU - Ruijie Zhang
AU - Huiteng Xu
AU - Shilin Guo
PY - 2026
DA - 2026/11/04
JO - Frontiers in Neurorobotics
DO - 10.3389/fnbot.2026.1967140
UR - https://doi.org/10.3389/fnbot.2026.1967140
AB - With the development of visual sensor technology and human pose estimation algorithms, skeleton-based action recognition has received extensive attention in sports analysis and training systems. The recognition of martial-arts actions is difficult because different styles can share similar poses while differing in rhythm and coordination. To support this task, we constructed the Martial Arts Pose (MA-Pose) dataset, which contains 433 source videos across 13 martial arts categories. The 33-landmark sequences include normalized image coordinates and monocular relative depth rather than metric 3D ground truth. We further propose a Dynamic Multi-Stream Multi-Scale Graph Convolutional Network (DMMS-GCN), which integrates joint, bone, and motion features. The model combines physical, learnable, and input-adaptive graph topology with parallel temporal convolutions. With a fixed seed of 42, 346 videos are used for training and 87 for validation, with one center crop and one prediction per validation video. DMMS-GCN correctly classifies 73/87 videos (83.91% Top-1; 83.51% Macro-F1; 95% Wilson confidence interval, 74.78%–90.17%) and retrieves the correct class among the top five for 84/87 videos (96.55%). These values describe development-set performance because model selection and evaluation use the same validation subset. On a Raspberry Pi 4B, pose extraction runs at approximately 10 frames/s and classification of a prepared 150-frame tensor takes approximately 120 ms. Collecting the first 150 processed pose frames takes about 15 s, so the deployment demonstrates edge-device feasibility rather than instantaneous recognition. The source code, model checkpoint, and MA-Pose dataset are publicly available.
SN - 1662-5218
ER -