两篇新的研究论文介绍了跨模态步态识别的新方法，旨在提高在二维摄像头和三维激光雷达等不同传感模态下识别个体的能力。DiffCrossGait 利用潜在扩散模型在根本上对齐步态轨迹，而不仅仅是最终的嵌入，从而促进模态不变的特征。TCFDNet 采用文本引导特征解耦，利用大型语言模型生成步态模式的语义描述，并对齐视觉和文本特征以实现更鲁棒的识别。 AI
影响 这些新技术可以通过在不同传感器输入之间实现更准确的识别来增强生物识别安全和监控系统。
排序理由 两篇在 arXiv 上发表的学术论文提出了跨模态步态识别的新方法。
AI 生成摘要 · Google Gemini · 来自 2 个来源。 我们如何撰写摘要 →
arXiv:2606.00153v1 Announce Type: cross Abstract: Cross-modal 2D-3D gait recognition is impeded by inherent domain discrepancies between 2D silhouette and 3D LiDAR range-view representations. While prior methods align only final embeddings, we propose DiffCrossGait, which reformu…
arXiv:2605.30784v1 Announce Type: new Abstract: Gait recognition is a biometric technique that identifies individuals based on their walking patterns, offering advantages in long-range, non-intrusive scenarios. However, real-world scenarios often involve heterogeneous sensing mod…