Event-Based Vision sensor characteristics
Scope
Event-based cameras and sensor arrays require specific calibration and the usual metrics of frame-based conventional cameras can't be straightforwardly applied to event-based cameras.
Various metrics have been defined on that purpose and some of them have been already used to characterized our first sensor generation [Sensitivity and uniformity of a 0.18Β΅m CMOS temporal contrast pixel array].
In this page we give a list the most recent Kpis, which are defined and used in our latest sensor version [GenX320].
The definition and method of theses key metrics depicted in the GenX320 datasheet, are also valid for our other sensors (Gen3.1, Gen4.0, Gen4.1, IMX636).
1 - Definition of light conditions
During experiments, event data are obtained using a warm-white LED light source with a color temperature of 3000 K.
Depending on the parameters to extract, contrast step or sinusoidal stimulation, in the form of rectangular or sine waves with controlled light levels and frequencies, or constant light levels, are applied to the pixel array.
Stimulation (luminance and illuminance) contrasts are defined according to Weberβs law:
πΆπππ‘πππ π‘ [%] =100Γ(πΏπππ₯ β πΏπππ)/πΏπππ
The light levels are referred to the focal plane of the event-based pixels.
2 - Key metrics
Those KPIs specifics to Event-Based camera are generally related to electro optical characteristics of the sensor:
a) Contrast Threshold (CT)
See GenX320 datasheet, page 34
b) Ramp Contrast Threshold (RCT)
c) Contrast Threshold Non-Uniformity (CTNU)
d) Ramp contrast threshold non-uniformity
e) Dynamic range
f) Pixel latency
g) Pixel bandwidth
h) Background rate
Note that the Background rate can be related as well to the notion of hot pixel tackled in this article: How to mask the effect of crazy/hot pixel
Also, background rate for Event-Based sensor could be considered as an equivalent of the SNR (Signal to Noise ratio) metric which traditionally provide the temporal (or dark) noise on frame-based sensors. This is used as a metric for understanding various performance capabilities of specific sensors in various lighting conditions.
As event sensors are tuned for the environment, traditional noise measurement metrics are irrelevant. You can check this link for more details.
As a Prophesee customer, get access to your personal ticketing tool, application notes, product manuals, SDK Download Center and more resources.