AUTHOR=Joubert Damien , Marcireau Alexandre , Ralph Nic , Jolley Andrew , van Schaik André , Cohen Gregory 

TITLE=Event Camera Simulator Improvements via Characterized Parameters

JOURNAL=Frontiers in Neuroscience

VOLUME=Volume 15 - 2021

YEAR=2021

URL=https://www.frontiersin.org/journals/neuroscience/articles/10.3389/fnins.2021.702765

DOI=10.3389/fnins.2021.702765

ISSN=1662-453X

ABSTRACT=It has been more than two decades since the first neuromorphic \ac{dvs} sensor was invented, and many subsequent prototypes have been built with a wide spectrum of applications in mind.  Competing against state-of-the-art neural networks in terms of accuracy is difficult, although there are clear opportunities to outperform conventional approaches in terms of power consumption and processing speed. As neuromorphic sensors generate sparse data at the focal plane itself, they are inherently energy-efficient, data-driven, and fast. It is these properties that need to be exploited to overcome the limitations of conventional approaches. In this work, we present an extended \ac{dvs} pixel model and simulator for neuromorphic systems and benchmarks, and provide validation and characterisation against a real sensor to reduce the gap between simulated and real data. Notably, our model avoids over-sampling the rendering engine, decreasing the computational cost, and modeling the essential features provided by the readout circuit. Using a dynamic variant of the MNIST dataset as a benchmarking task, we use this simulator to explore how the latency of the sensor can be used to outperform conventional sensors in terms of sensing speed. Better exploitation of the benefits offered by neuromorphic sensors requires a thorough exploration of a more realistic sensor simulation on dynamic tasks. This paper presents a step toward that goal by providing a framework for a more holistic comparison of the simulated sensors' performance on a reading task specially developed for this purpose.