Note

If you acquired a sensor or an EVK from Prophesee, you can find datasheets and user manuals in our Knowledge Center once you will have requested an account. In this page, we give complementary information about sensor’s biases.

Biases

Prophesee’s sensors provide a set of adjustable settings, which allow the sensor performance to be tuned for different application requirements and conditions (higher speed, lower background activity, higher contrast sensitivity threshold, etc.). These sensor settings are called biases.

Note

In order to properly understand how biases can affect the behaviour of the sensors, you may want to watch our Bias Tuning and Region of Interest video and also Training Video on Event-Based Vision Sensor as it includes an overview of concepts mentioned in this page like contrast sensitivity, background noise and pixel latency.

Available Biases

There are five generally useful biases:

• Contrast sensitivity threshold biases (bias_diff_on, bias_diff_off)

  The pixels of an event sensor are triggered by a change in illumination of a certain contrast. The minimum contrast to produce an event, measured in %, is called the contrast threshold. These biases control the contrast threshold - the sensitivity of the sensor to contrast changes.

  • bias_diff_on: adjusts the contrast threshold for ON events. This determines the ON contrast threshold, the factor by which the pixel must get brighter, before an ON event occurs for that pixel. It is usually adjusted when the user wants to change how many events are output during a big change in illumination, or else to change the sensitivity to small positive light changes.

  • bias_diff_off: adjusts the contrast threshold for OFF events. This determines the OFF contrast threshold, the factor by which the pixel must get darker, before an OFF event occurs for that pixel. It is usually adjusted when the user wants to change how many events are output during a big change in illumination, or else to change the sensitivity to small negative light changes.

Note

There is no symmetry between bias_diff_off and bias_diff_on. In other words, setting those two bias parameters to the same value will not necessary lead to the same number of ON and OFF events generated with a dedicated light stimulus.

This is mostly due to the fact that the Contrast Detection Probability (CDP) of an ON event is different from the CDP of an OFF event. For more details about CDP, please refer to the corresponding section in this KC article.

• Bandwidth biases (bias_fo, bias_hpf) control the pixel low-pass and high-pass filters.

  • bias_fo: adjusts the low-pass filter. This changes how rapidly fluctuating light is filtered out. It determines the maximum rate of change of illumination that can be detected. It is often used to remove flickering and noise in the scene, but doing so will also increase the latency of the sensor.

  • bias_hpf: adjusts the high-pass filter. This determines how slow illumination change is filtered out. It determines the minimum rate the light must change, for the pixel to output an event. It is often used to remove the background of the scene and show only fast moving objects, or to reduce the background noise.

• Dead time bias (bias_refr) controls the pixel “refractory period”.

  • bias_refr: adjusts the refractory period. This determines the duration for which the pixel is blind after each event. It can be used to change the number of events during a big illumination change, without changing the sensitivity or the bandwidth. It is often used to make each big light change produce only one event. To get the mapping of this bias to microseconds, please refer to your sensor datasheet.

All biases affect both the range of signals which can be detected, and the rate of noise background events observed. Reducing the noise is always possible, but also reduces in some way the range of useful signals which can be detected.

Note

There is an extra bias (bias_pr) that drives the photoreceptor bandwidth by controlling the front-end part of the pixel. We don’t recommend changing this bias, except in very specific conditions. For Gen4.1 and IMX636 sensors, this bias is not available anymore (neither via API nor in our players).

Since the Metavision SDK 4.0 release, a new method called get_pixel_dead_time has been added to I_Monitoring to retrieve pixel dead time (aka. “refractory period”). Also, this function can allow to get the mapping of bias_refr to microseconds for your Gen4.1 or IMX636 sensor.

Bias Tuning

The different generations of Prophesee sensors use different units and scales to configure the biases. For example, a given biases configuration for a Gen3.1 can not be used on an Gen4.1. Even the default biases configuration can not really be considered equivalent from one generation to another because the pixel and overall sensor technology and architecture varies.

The main differences are:

• Between Gen3.1 and Gen4.0/Gen4.1/IMX636, some biases are expressed in opposite ways

• In IMX636, biases are expressed as relative offsets

In the following table we describe how biases should be tuned depending on the sensor generation:

Bias Name	Gen3, Gen3.1	Gen4, Gen4.1	IMX636
All biases	All the biases are expressed in mV	All the biases are expressed as a positive digital number	All the biases are expressed as a positive or negative digital number, as relative offsets around the default value. The default bias is always 0
bias_diff_on	The contrast threshold is defined by the difference of bias_diff_on above bias_diff. To increase the contrast threshold, increase bias_diff_on.	The contrast threshold is defined by the difference of bias_diff_on above bias_diff. To increase the contrast threshold, increase bias_diff_on.	The contrast threshold is defined only by the bias_diff_on setting. To increase the contrast threshold, increase bias_diff_on.
bias_diff_off	The contrast threshold is defined by the difference of bias_diff_off below bias_diff. To increase the contrast threshold, decrease bias_diff_off.	The contrast threshold is defined by the difference of bias_diff_off below bias_diff. To increase the contrast threshold, decrease bias_diff_off.	The contrast threshold is defined only by the bias_diff_off setting. To increase the contrast threshold, increase bias_diff_off.
bias_fo	Decrease bias_fo to widen the bandwidth	Increase bias_fo to widen the bandwidth Note: in Gen4 bias_fo is called bias_fo_n	Increase bias_fo to widen the bandwidth
bias_hpf	Increase bias_hpf to widen the bandwidth and make the sensor sensitive to slower light changes.	Decrease bias_hpf to widen the bandwidth and make the sensor sensitive to slower light changes.	Decrease bias_hpf to widen the bandwidth and make the sensor sensitive to slower light changes.
bias_refr	Decrease bias_refr to decrease the refractory period after each event	Increase bias_refr to decrease the refractory period after each event	Increase bias_refr to decrease the refractory period after each event

Note

For IMX636, the rationale behind the new bias values management is that default values for the biases may vary slightly from sensor to sensor depending on factory trimming. In order for a given sensor tuning to result in the closest possible performance when applied to another sensor, the user-tunable values in our API and tools are now an offset relative to the default bias values.

However, it is possible to read absolute bias values of a given sensor. First, you need to set the logging level to Trace. Then, launch Metavision Studio in the same terminal window. Now, when setting a bias parameter inside studio, the absolute value of that bias parameter will be displayed in the terminal. Below is an example of output after setting bias_fo and bias_hpf from 0 to 8 and 12 respectively:

[SERVER] - stderr - [SDK][STUDIO] Building from: {"type":"setBias","payload":{"bias":"bias_fo","value":8},"meta":{"scope":"local"}} size: 82 cmd: setBias
[HAL][TRACE] register name: bias_fo , factory default: 81 , current value: 81 , diff: 0 , value range: [ -35 ,  55 ]
[HAL][TRACE] register name: bias_fo , factory default: 81 , current value: 89 , diff: 8 , value range: [ -35 ,  55 ]

[SERVER] - stderr - [SDK][STUDIO] Building from: {"type":"setBias","payload":{"bias":"bias_hpf","value":12},"meta":{"scope":"local"}} size: 84 cmd: setBias
[HAL][TRACE] register name: bias_hpf , factory default: 0 , current value: 0 , diff: 0 , value range: [ 0 ,  120 ]
[HAL][TRACE] register name: bias_hpf , factory default: 0 , current value: 12 , diff: 12 , value range: [ 0 ,  120 ]

There is no one fits all approach to tune the biases as it depends on multiple parameters (e.g. objects in the scene, light condition etc.). Nevertheless, those steps are considered valid in many scenarios are a good starting point:

• first update bias_fo to reduce impacts of rapidly fluctuating light (high frequencies),

• then update bias_hpf to reduce background noise (low frequencies),

• finally you can adjust bias_diff_on and bias_diff_off to adapt the Contrast Sensitivity to your application.

To tune the biases, the easiest way is to use Metavision Studio where the Biases panel available in the Settings allows to adjust the biases and immediately see the effect on the display window:

To get more information on the biases consult the Application Note on bias tuning in Knowledge Center, where we describe the visual effect on tuning the sensor biases.

Bias Ranges

Not all bias values are allowed. There are specific acceptable ranges depending on your sensor’s generation. Those ranges are enforced by each camera plugin through the HAL API, so when trying to set out of range values you will get warning message similar to [HAL][WARNING] 100 is not in acceptable range for bias_fo

Bias ranges for Gen3.1 sensor:

Bias Name	Default value	Minimum value	Maximum value	Comment
bias_diff	299	200	400	recommended to not change
bias_diff_on	384	bias_diff + 75	bias_diff + 200	if bias_fo < 1350, Min Value = bias_diff + 95
bias_diff_off	222	100	bias_diff - 65	if bias_fo < 1350, Max Value = bias_diff - 85
bias_fo	1477	1250	1800
bias_hpf	1499	900	1800
bias_refr	1500	1300	1800

Note

For Gen3.1, note that not all values of biases can be configured within the sensor. For example, setting a bias_hpf of 1445 or 1455 will in fact set the bias to the actual value of 1448. This is not visible in Metavision Studio, but you will notice it when saving the bias values into BIAS files (see below).

Bias ranges for Gen4.1 sensor:

Bias Name	Default value	Minimum value	Maximum value	Comment
bias_diff	80	52	100	recommended to not change
bias_diff_on	115	bias_diff + 15	140
bias_diff_off	52	25	bias_diff - 15
bias_fo	74	45	110
bias_hpf	0	0	120
bias_refr	68	30	100

Bias ranges for IMX636 sensor:

Bias Name	Default value	Minimum value	Maximum value	Comment
bias_diff	0	-32	16	recommended to not change
bias_diff_on	0	-85	140
bias_diff_off	0	-35	190
bias_fo	0	-35	55
bias_hpf	0	0	120
bias_refr	0	-20	235

Warning

Not respecting those ranges may lead to sensor malfunction or damage. If you think your application requires using out of range values for some biases, you can disable the bias ranges check when starting your device (this option is available in Metavision Studio or through the HAL API using Metavision::DeviceConfig class. Beware that this setting is for advanced users and that we don’t guarantee the behavior of the sensors when using out of ranges biases values. In case of doubt, please contact us to discuss this topic.

Bias Files

SDK allows to store the bias values in files with a .bias extension. These BIAS files are handled in Metavision Studio, Metavision Viewer and multiple SDK Code Samples using the C++ class Metavision::Biases.

The format of the BIAS files is the following:

• the file is a text file where each line configures a specific bias

• each line is made of bias_value and bias_name separated by percentage sign (%).

For example an IMX636 bias file could have this content:

0 % bias_diff
10 % bias_diff_off
10 % bias_diff_on
-5 % bias_fo
0 % bias_hpf
0 % bias_refr

After installing Metavision SDK, BIAS files storing default bias values for every sensor generation can be found in <install-prefix>/share/metavision/hal/resources/biases/.

Biases API

Biases can be tuned using HAL and SDK Driver API:

• HAL C++ API: Metavision::I_LL_Biases facility

• HAL Python API: metavision_hal.I_LL_Biases

• SDK Driver C++ API: Metavision::Biases class