Event

Penn Develops General-Purpose Event Camera Representation Tool F³

1 reports · 1 sources

A University of Pennsylvania engineering team has proposed a general-purpose computational tool, Fast Feature Field (F³), that converts the tens of millions of signals per second from event cameras into representations usable by existing computer vision software; the paper was published in Science Robotics. The team says methods based on F³ achieve state-of-the-art results on multiple robot perception tasks, with algorithms about 2 to 5 times faster than previous methods; F³ itself processes HD event camera data at 120 times per second, and at the lower VGA resolution at 440 times per second. The team tested F³ on data collected from cars, quadruped robots and drones, with tasks covering object recognition, distance estimation and motion tracking, in scenarios including nighttime city streets and crowded indoors. The team is using F³ to build autonomous drones capable of navigating in cluttered, low-light environments.