DiffCrossGait: Trajectory-Level Alignment for 2D-3D Cross-Modal Gait Recognition via Latent Diffusion
Two new research papers introduce novel approaches to cross-modal gait recognition, aiming to improve the identification of individuals across different sensing modalities like 2D cameras and 3D LiDAR. DiffCrossGait utilizes latent diffusion models to align gait trajectories at a fundamental level, rather than just final embeddings, promoting modality-invariant features. TCFDNet employs text-guided feature disentanglement, using large language models to generate semantic descriptions of gait patterns and align visual and textual features for more robust recognition. AI
IMPACT These new techniques could enhance biometric security and surveillance systems by enabling more accurate identification across diverse sensor inputs.