Event-based real-time optical flow estimation

In this paper, we introduce a novel approach defining features and estimating their optical flow with an event-based sensor. Event cameras detect temporal intensity changes on low latency with high dynamic range, arising as a promising solution for improving Visual odometry (VO). However, it produces an event stream rather than a framewise image, requiring additional procedures like rebuilding intensity images to apply frame based algorithms. Hence, some information degenerates to process event data with conventional algorithms. Therefore, there are needs on event-based algorithms to fully utilize the low latency characteristics of event cameras. We propose an algorithm to extract features and estimate its optical flow only with event stream, empowering the event camera on VO.