Probabilistic Approach for Detection of High-Frequency Periodic Signals using an Event Camera
David El-Chai Ben-Ezra, Ron Arad, Ayelet Padowicz, Israel Tugendhaft
TL;DR
The paper addresses detecting high-frequency pixel-sized periodic signals with asynchronous event cameras by formulating a per-pixel probabilistic detection criterion based on inter-event timing. It introduces an asynchronous time-surface–driven algorithm that accumulates per-pixel statistics and makes on-the-fly detections when the per-pixel false-alarm probability, approximated by $Q(m,d)$, falls below a threshold $q$; the surrogate relies on an exponential inter-arrival model with $P(m)$ and explicit expressions for $Q(m,d)$. Theoretical analysis connects the practical algorithm to a new order-statistics–flavored problem, provides a tractable approximation to the false-alarm probability, and demonstrates robust detection of a $100$ Hz streetlight flicker in urban scenes, highlighting potential for high-frequency sensing and flicker removal in event-based vision.
Abstract
Being inspired by the biological eye, event camera is a novel asynchronous technology that pose a paradigm shift in acquisition of visual information. This paradigm enables event cameras to capture pixel-size fast motions much more naturally compared to classical cameras. In this paper we present a new asynchronous event-driven algorithm for detection of high-frequency pixel-size periodic signals using an event camera. Development of such new algorithms, to efficiently process the asynchronous information of event cameras, is essential and being a main challenge in the research community, in order to utilize its special properties and potential. It turns out that this algorithm, that was developed in order to satisfy the new paradigm, is related to an untreated theoretical problem in probability: let $0\leqτ_{1}\leqτ_{2}\leq\cdots\leqτ_{m}\leq1$, originated from an unknown distribution. Let also $ε,δ\in\mathbb{R}$, and $d\in\mathbb{N}$. What can be said about the probability $Φ(m,d)$ of having more than $d$ adjacent $τ_{i}$-s pairs that the distance between them is $δ$, up to an error $ε$ ? This problem, that reminds the area of order statistic, shows how the new visualization paradigm is also an opportunity to develop new areas and problems in mathematics.