Live Demonstration: Tangentially Elongated Gaussian Belief Propagation for Event-Based Incremental Optical Flow Estimation

Abstract

Optical flow estimation is a fundamental functionality in computer vision. An event-based camera, which asynchronously detects sparse intensity changes, is an ideal device for realizing low-latency estimation of the optical flow owing to its low-latency sensing mechanism. We developed an efficient full-flow estimation called Tangentially elongated Gaussian belief propagation (TEGBP). TEGBP formulates the full flow estimation as the marginalization of probability using a message-passing based on the BP. The formulation permits event-by-event asynchronous incremental updates of the full flow; i.e., given a normal-flow observation, it updates its belief about full flow by asynchronous local communication. This paper presents a OpenMP based real-time full-flow estimation demo by taking advantage of the asynchronous formulation. Specifically, we parallelize the individual sequence of the message exchange evoked by a single normal-flow observation. Beliefs at each node are updated on an event-by-event basis manner in parallel, realizing the real-time procession on CPUs. Our C++ code is available at https://github.com/DensoITLab/tegbp.