Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Quantifying Noise of Dynamic Vision Sensor

Evgeny V. Votyakov, Alessandro Artusi

TL;DR

This work addresses the challenge of background activity (BA) noise in Dynamic Vision Sensors (DVS) by proposing an objective denoising criterion based on Detrended Fluctuation Analysis (DFA). The authors adapt DFA to DVS time series to quantify long-range correlations and use this as a metric to evaluate BA filtering without ground truth, deriving a comparison between clean and noisy components. They demonstrate the approach on a moving-car dataset with a simple BA filter, showing that increasing the BA filter window parameter $\Delta T$ can yield higher SNR and noise that behaves more like uncorrelated noise (\(\alpha\approx 0.5\)). The work offers a practical, ground-truth-free method to guide BA filter design and parameter tuning, with future directions including tests on more complex scenes and connections between DFA exponents and event-time statistics.

Abstract

Dynamic visual sensors (DVS) are characterized by a large amount of background activity (BA) noise, which it is mixed with the original (cleaned) sensor signal. The dynamic nature of the signal and the absence in practical application of the ground truth, it clearly makes difficult to distinguish between noise and the cleaned sensor signals using standard image processing techniques. In this letter, a new technique is presented to characterise BA noise derived from the Detrended Fluctuation Analysis (DFA). The proposed technique can be used to address an existing DVS issues, which is how to quantitatively characterised noise and signal without ground truth, and how to derive an optimal denoising filter parameters. The solution of the latter problem is demonstrated for the popular real moving-car dataset.

Quantifying Noise of Dynamic Vision Sensor

TL;DR

This work addresses the challenge of background activity (BA) noise in Dynamic Vision Sensors (DVS) by proposing an objective denoising criterion based on Detrended Fluctuation Analysis (DFA). The authors adapt DFA to DVS time series to quantify long-range correlations and use this as a metric to evaluate BA filtering without ground truth, deriving a comparison between clean and noisy components. They demonstrate the approach on a moving-car dataset with a simple BA filter, showing that increasing the BA filter window parameter can yield higher SNR and noise that behaves more like uncorrelated noise (). The work offers a practical, ground-truth-free method to guide BA filter design and parameter tuning, with future directions including tests on more complex scenes and connections between DFA exponents and event-time statistics.

Abstract

Dynamic visual sensors (DVS) are characterized by a large amount of background activity (BA) noise, which it is mixed with the original (cleaned) sensor signal. The dynamic nature of the signal and the absence in practical application of the ground truth, it clearly makes difficult to distinguish between noise and the cleaned sensor signals using standard image processing techniques. In this letter, a new technique is presented to characterise BA noise derived from the Detrended Fluctuation Analysis (DFA). The proposed technique can be used to address an existing DVS issues, which is how to quantitatively characterised noise and signal without ground truth, and how to derive an optimal denoising filter parameters. The solution of the latter problem is demonstrated for the popular real moving-car dataset.
Paper Structure (8 sections, 8 equations, 4 figures)

This paper contains 8 sections, 8 equations, 4 figures.

Table of Contents

  1. Introduction
  2. Dynamic Visual Sensor (DVS)
  3. Background Activity (BA) filter
  4. Formulation of the mathematical problem
  5. Detrended Fluctuation Analysis
  6. Results
  7. Conclusion
  8. Acknowledgment

Figures (4)

  • Figure 1: (a) clean $(TS)_\text{clean}$ and noise $(TS)_\text{noise}$ are produced by a BA filter (\ref{['eq:BAFdefintion']}) from an initial $(TS)_\text{full}$. (b) DFA results. Slope of the line is DFA scaling exponent $\alpha$ characterizing an impact of long term correlations.
  • Figure 2: BAF applied at various $\Delta T$.
  • Figure 3: DFA applied to the noise time series.
  • Figure 4: SNR and $\alpha$ as a function of $\Delta T$.