Computer Vision
融合步态变换器
FusionGait-transformerA multimodal transformer-based framework that fuses pose, silhouette, and inertial time series data using self-attention and cross-attention mechanisms to achieve robust gait recognition and gait-condition classification under occlusion and domain shift. / 一种基于多模态变换器的框架，通过自注意力和交叉注意力机制融合姿态、轮廓和惯性时间序列数据，以在遮挡和域偏移下实现鲁棒的步态识别和步态条件分类。
- Knowledge type
- 实际应用
- First proposed or discovered
- 2026
- Proposer
- Kiran Kumar S. V. N. Madupu, Jagadeeshwar Madiraju, Naren Kodali, Soma Sekhar Goviraboyina
- Creator
- 农场主
- Last updated
- 2026-10-31 19:06:53
Sub-knowledge points
Per-modality reliability gateMasked multimodal pretrainingDomain adversarial training
Higher-order knowledge points
Gait recognitionGait-condition classificationWearable healthcare settings
Main parameters and performance indicators
Rank-1 identification accuracy: 98.9%Equal error rate (EER): 1.8%Gait-condition classification F1-score: 95.7%Area under the ROC curve (AUC-ROC): 97.6%
Open problems
Validation with real wearable sensors and clinically labeled cohortsPerformance under uncontrolled benchmark conditions
Related literature
ranked by community starsFusionGait-transformer: multimodal transformer-based framework for robust gait recognition and gait-condition classificationOpen source
Related products and links
FusionGait-transformer framework
Creation and modification trail
1 recordsRecord title: FusionGait-transformer.
Node update feedback
review required