Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2509.08661 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866918143266390016 |
|---|---|
| author | Liu, Liangjin Zheng, Haoyang Zhu, Zhengzhong Zhou, Pei |
| author_facet | Liu, Liangjin Zheng, Haoyang Zhu, Zhengzhong Zhou, Pei |
| contents | Isolated Sign Language Recognition (ISLR) is challenged by gestures that are morphologically similar yet semantically distinct, a problem rooted in the complex interplay between hand shape and motion trajectory. Existing methods, often relying on a single reference frame, struggle to resolve this geometric ambiguity. This paper introduces Dual-SignLanguageNet (DSLNet), a dual-reference, dual-stream architecture that decouples and models gesture morphology and trajectory in separate, complementary coordinate systems. The architecture processes these streams through specialized networks: a topology-aware graph convolution models the view-invariant shape from a wrist-centric frame, while a Finsler geometry-based encoder captures the context-aware trajectory from a facial-centric frame. These features are then integrated via a geometry-driven optimal transport fusion mechanism. DSLNet sets a new state-of-the-art, achieving 93.70%, 89.97%, and 99.79% accuracy on the challenging WLASL-100, WLASL-300, and LSA64 datasets, respectively, with significantly fewer parameters than competing models. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2509_08661 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Skeleton-based sign language recognition using a dual-stream spatio-temporal dynamic graph convolutional network Liu, Liangjin Zheng, Haoyang Zhu, Zhengzhong Zhou, Pei Computer Vision and Pattern Recognition Artificial Intelligence I.2.m; I.2.0 Isolated Sign Language Recognition (ISLR) is challenged by gestures that are morphologically similar yet semantically distinct, a problem rooted in the complex interplay between hand shape and motion trajectory. Existing methods, often relying on a single reference frame, struggle to resolve this geometric ambiguity. This paper introduces Dual-SignLanguageNet (DSLNet), a dual-reference, dual-stream architecture that decouples and models gesture morphology and trajectory in separate, complementary coordinate systems. The architecture processes these streams through specialized networks: a topology-aware graph convolution models the view-invariant shape from a wrist-centric frame, while a Finsler geometry-based encoder captures the context-aware trajectory from a facial-centric frame. These features are then integrated via a geometry-driven optimal transport fusion mechanism. DSLNet sets a new state-of-the-art, achieving 93.70%, 89.97%, and 99.79% accuracy on the challenging WLASL-100, WLASL-300, and LSA64 datasets, respectively, with significantly fewer parameters than competing models. |
| title | Skeleton-based sign language recognition using a dual-stream spatio-temporal dynamic graph convolutional network |
| topic | Computer Vision and Pattern Recognition Artificial Intelligence I.2.m; I.2.0 |
| url | https://arxiv.org/abs/2509.08661 |