EventGait: Towards Robust Gait Recognition with Event Streams
Researchers have developed EventGait, a novel dual-stream framework for gait recognition using event cameras. This approach processes motion and shape information separately, leveraging a Mixture of Spiking Experts for dynamic perception and Cross-modal Structure Alignment for shape representation. To facilitate research, they also introduced two new benchmarks, SUSTech1K-E and CCGR-Mini-E, and a synthesis pipeline for event-based gait data. EventGait demonstrates superior performance in low-light conditions compared to traditional camera-based methods and sets a new state-of-the-art on event-based gait recognition benchmarks. AI
IMPACT Enhances biometric security by enabling robust identification in challenging low-light environments.