ANNOUNCEMENT: Ph.D. Dissertation Defense
Name: Andrew Berkovich
Professor Pamela Abshire, Chair/Advisor
Professor Timothy K. Horiuchi
Professor Martin Peckerar
Dr. Geoffrey L. Barrows, Centeye Inc.
Professor Yiannis Aloimonos, Dean's Representative
Date/Time: Wednesday, May 3, 2017 / 10am
Place: Room 1146 AVW Building
Title: Biologically-Inspired Low-Light Vision Systems for Micro-Air Vehicle Applications
Abstract: The field of small and micro air vehicles (MAVs) has grown substantially in the past decade, with proven applications for military and increasingly civilian use. However there are two scenarios in which such air vehicles are still, as a common practice, not being used – that of flying outdoors at night, and that of flying indoors in an unlit environment. Both applications are similar in that they involve operations in photon limited environments, e.g. at extremely low light levels. This is particularly the case if the air vehicle is being flown deep inside a building or a cave, or deep underneath a forest canopy at night time. In such environments, ambient light levels can drop to the point that image sensors would have to acquire useful images with just thousands (or fewer) photons per second. This basic capability is currently nowhere close to being met.
Biology offers ample evidence that visual navigation in low light environments is possible, as there are several animals that routinely fly in extremely low light environments in which each photoreceptor receives just several photons per second. These animals present a clear existence proof that it is possible to navigate visually at such low light levels. Moreover, they provide design inspiration for biologically-inspired low-light vision hardware.
This talk highlights biologically-inspired vision systems that extend the operating range at which MAVs can self-stabilize, perceive, and in general operate at low light levels. We will discuss custom vision hardware built upon spiking, single-photon avalanche diodes (SPADs) for imaging in the visible spectrum, noise reduction techniques with adaptive analog circuits for room-temperature short wave infrared (SWIR) sensing, and benchmarking results for optic flow processing at low-light levels.