LC-Flow: Learning Local Continuous Optical Flow and Confidence from events
Two new research papers propose novel methods for estimating optical flow using event-based cameras. LC-Flow introduces a recurrent neural network that maintains temporal continuity by accumulating event data, addressing limitations of frame-based and stateless methods. The second paper, From Contrast to Consistency, presents a hybrid-supervised framework that emphasizes spatio-temporal structural consistency and trajectory continuity, overcoming challenges with limited ground-truth data and improving motion coherence. AI
IMPACT These papers advance the state-of-the-art in event-based vision, potentially improving real-time motion analysis for robotics and autonomous systems.