University of Pennsylvania Develops General-Purpose Computing Tool Called 'Fast Feature Fields'

Engineers at the University of Pennsylvania have developed a general-purpose computational tool called Fast Feature Field (F3). The tool can convert the tens of millions of signals generated by event cameras per second into data information usable by robots. The related paper was published in the journal Science Robotics. The senior author is Pratik Chaudhari, Associate Professor in the Department of Electrical and Systems Engineering and a secondary faculty member in the Department of Computer and Information Science. Co-authors include Kostas Daniilidis, Ruth Yalom Stone Professor in the Department of Computer and Information Science, and the first author is Richeek Das, a doctoral student in the Department of Computer and Information Science. Event cameras work by continuously monitoring changes in light intensity, in contrast to traditional cameras, which record full images at fixed time intervals (for example, 30 frames per second). Previously, processing event camera data required relying on a combination of multiple specialized algorithms, each typically designed for only a single task. The F3 tool enables event cameras to be applied on platforms such as autonomous vehicles and drones for object recognition, distance estimation, and motion perception.