Bias & Tuning

Pixel behavior in an event camera is decided by a set of configurable parameters collectively called bias. Understanding bias is the key to tuning an event camera to "see clearly, with low noise". This page covers the general concepts; the specific tunable parameters and values are on each sensor's product page (placeholder, to be filled).

What is bias

Each pixel of an event camera independently judges whether "a brightness change exceeds the threshold before firing an event". Bias is the set of parameters that controls this judgment, commonly including:

  • Sensitivity / contrast threshold: how large the brightness change must be to trigger an event.
  • Refractory / dead time: how long after firing an event a pixel must wait before firing again.
  • Bandwidth: the response speed of the pixel's front-end circuit.

The bias items differ in naming and count across sensors; refer to the parameter list of the corresponding sensor.

Tuning trade-offs

Tuning is fundamentally about finding balance between two extremes:

  • High sensitivity (low threshold) → captures faint changes, but background activity noise also increases and the picture looks "dirty".
  • Low sensitivity (high threshold) → less noise and a cleaner picture, but may miss fast or faint targets.

Heuristics

For static or low-light scenes you can raise sensitivity slightly to see the target; for bright light or high-background-activity scenes, lower the sensitivity to suppress noise. Combined with a denoising algorithm (see Algo), you can keep sensitivity high while pressing noise down.

Relationship with accumulation time

Bias decides "whether to fire events", and accumulation time decides "how events are accumulated into a picture". The two must be paired:

  • High bias + short accumulation → an all-black picture (no events to accumulate).
  • Low bias + long accumulation → a smeared picture (too many events stacked).

See Fundamentals for accumulation time.

ROI (Region of Interest)

You can enable event output for only some pixels in the frame (ROI), reducing data volume, bandwidth, and compute. This suits scenarios where the target position is fixed (e.g. production-line inspection, eye tracking).

Bias parameters for our sensors

Warning

Bias parameters differ across sensors; this page covers only general concepts. The specific parameter lists and default values will be filled in after the hardware team supplies them; before tuning, refer to the specification of the corresponding sensor — do not directly apply parameters from other models.