TPR Seminar - Continuous-time, Event-based Vision

Cedric Scheerlinck

Abstract:

Event cameras provide asynchronous, data-driven measurements of pixel-wise temporal contrast with extremely high temporal resolution over a large dynamic range. Events represent discrete pixel-wise brightness changes, and they are transmitted almost instantaneously. A key advantage over conventional cameras is that event cameras only transmit non-redundant information, rather than synchronously updating the full image frame at an arbitrary rate. This enables efficient local processing of salient image regions. I propose to estimate a continuous-time image state that is updated asynchronously based on events. Image gradient, which is important in motion perception, feature extraction and edge-detection can also be estimated from events. Finally I propose to estimate a continuous-time motion state that describes 2D optical flow, based on estimated image gradient and the precise timing of events.

Biography

Cedric completed my Master of Engineering (Mechanical) at the University of Melbourne in 2016. In 2015 he worked as a research assistant in the Fluid Dynamics lab at Melbourne before completing an exchange semester at ETH Zurich. His final year Master's thesis was a combined project with the University of Melbourne and The Northern Hospital, performing computational fluid dynamics studies on patient-specific coronary arteries.

In 2017 Cedric commenced hisPhD in Robotic Vision under the supervision of Rob Mahony at the ANU. His PhD topic is the development of novel optical flow algorithms capable of running in real-time for high-speed robotics applications. He am currently pursuing this research using event cameras, which are bio-inspired vision sensors with microsecond temporal resolution.