Why event cameras could finally make gait recognition work in the dark

Gait recognition (identifying people by how they walk) fails in poor lighting with standard cameras, but event cameras—which detect microsecond changes in brightness—naturally excel there. EventGait uses a dual-stream approach: one network with spiking neurons captures fine motion dynamics, another learns shape from vision models. The team released two new event-camera gait datasets and showed their method outperforms traditional cameras in low-light conditions while matching them in normal light. Code and data are available.