Flow‐assisted visual tracking using event cameras

The data from event cameras not only portray contours of moving objects but also contain motion information inherently. Herein, can be used in event-based and frame-based object trackers to ease the challenges occluded association, respectively. In tracker, events within a short interval are accumulated. Within interval, histogram local time measurements (or ‘motion histogram’) is proposed as feature describe target candidate regions. Then mean-shift tracking approach by shifting tracker towards similarity maximisation on histograms between As for given assumption that single moves at constant velocity image plane, distribution timestamps modelled, followed which object-level velocities obtained parameter estimation. We then build Kalman-based ensemble, deemed an additional measurement top detection results. Experiments have been conducted measure performance based our self-collected data. Thanks assistance information, successfully differentiates partially overlapped with distinct profiles; inter-frame avoids association failure fast-moving leads fast convergence