Event-Enhanced Hand Behavior Modeling: Toward Robust Dexterous Hand Motion Transfer

Transferring human hand skills to robots is critical for biometric behavior analysis and human-robot interaction. Existing methods primarily rely on RGB cameras, which suffer from motion blur and overexposure conditions, limiting their applicability in real-world scenarios. To address this, we propose a novel framework that synergistically combines event cameras (bio-inspired sensors with high dynamic range and low latency) with RGB data, enabling robust hand imitation through spatiotemporal synchronization. Our approach leverages the complementary strengths of event streams (capturing fast motion details) and RGB images (providing texture and edge information), effectively mitigating challenges like overexposure and motion blur. Furthermore, we contribute a large-scale synthetic event dataset tailored for dexterous hand analysis, supporting future research in biometric behavior modeling and dexterous motion transfer. Experiments demonstrate improvements over RGB-only methods, particularly in overexposure and fast motion scenarios, with enhanced accuracy and stability in both synthetic and real-world evaluations. This work bridges the gap between bio-inspired sensing, behavioral biometrics, and robots.