D405 Setting Exposure in OpenCV

[DaveBGld]

Using cap.set(cv2.CAP_PROP_EXPOSURE, exposureValue) no matter what exposure value (i.e. -4, 100000) does nothing.

I have had to resort to this code:

pipeline = rs.pipeline()
config = rs.config()
config.enable_stream(rs.stream.color, 1280, 720, rs.format.bgr8, 30)
pipeline.start(config)
depth_sensor = pipeline.get_active_profile().get_device().first_depth_sensor()
exposureValue = 100000
depth_sensor.set_option(rs.option.exposure, exposureValue)
pipeline.stop()

camera_index = 4
cap = cv2.VideoCapture(camera_index)

But this code slows the fps to 6 instead of 30....

Specific question: What is the preferred way to set the exposure value in OpenCV Python?

Thanks!

DBG

[MartyG-RealSense]

Hi @DaveBGld Your FPS is lagging because of the '100000' exposure value that you have set. When setting a manual exposure value, certain value ranges can cause FPS to slow, as described by a RealSense team member at #1957 (comment)

The default exposure value for the D405 is '33000'.

The D405 does not have a separate RGB sensor and so shares its exposure setting with depth. So whatever the depth exposure is manually set to will affect RGB color exposure too. I confirmed in my own tests that FPS is 30 at exposure 33000 and drops to 6 at 100000 exposure.

OpenCV has an exposure command called CAP_PROP_EXPOSURE for altering image brightness, though I note that you have already tried it. The links below have Python examples for implementing it.

https://www.kurokesu.com/main/2020/05/22/uvc-camera-exposure-timing-in-opencv/

https://www.principiaprogramatica.com/2017/06/11/setting-manual-exposure-in-opencv/

opencv/opencv#9738 has advice from others whom CAP_PROP_EXPOSURE did not work for. According to their comments, CAP_PROP_AUTO_EXPOSURE should also be set to '0.25' in order for CAP_PROP_EXPOSURE to work.

[DaveBGld]

@MartyG-RealSense Thanks for the reply

I tried this:

cap = cv2.VideoCapture(camera_index)
cap.set(cv2.CAP_PROP_AUTO_EXPOSURE, 0.25) # manual mode
cap.set(cv2.CAP_PROP_EXPOSURE, -10) # This should return a black frame

And also

cap.set(cv2.CAP_PROP_AUTO_EXPOSURE, 0.75) # auto mode

before setting manual mode

And also using values 3 and 1 for auto and manual exposure.

Nothing works.

The OpenCV forums point out that these values are back-end (Intel libreaalsense) dependent. People report different values for different cameras...

[MartyG-RealSense]

The 'native' backend of librealsense is V4L2 by default (RSUSB = false). With the native backend, RealSense camera streams are accessible to standard non-RealSense Linux tools.

[DaveBGld]

@MartyG-RealSense thanks for that.

Still the code you suggested

cap = cv2.VideoCapture(camera_index)
cap.set(cv2.CAP_PROP_AUTO_EXPOSURE, 0.25) # manual mode
cap.set(cv2.CAP_PROP_EXPOSURE, -10) # This should return a black frame

doesn't appear to have any effect on the camera feed.

Any other ideas?

Thanks,

DBG

[MartyG-RealSense]

What if you set VideoCapture as '0' instead of camera_index?

cap = cv2.VideoCapture(0)

[DaveBGld]

@MartyG-RealSense thanks, but I have two cameras connected

[MartyG-RealSense]

I do not have any other suggestions to offer, unfortunately. I do apologize.

[DaveBGld]

So the D405 camera is not compatible with OpenCV...

[MartyG-RealSense]

RealSense cameras are fully compatible with OpenCV.

https://github.com/IntelRealSense/librealsense/tree/master/wrappers/opencv

https://github.com/IntelRealSense/librealsense/blob/master/doc/stepbystep/getting_started_with_openCV.md

Typically, auto-exposure is used with RealSense OpenCV applications rather than manual exposure though.

[DaveBGld]

If it was fully compatible, my project would not be struggling and stuck on this. :(

I guess the next step is getting some of the OpenCV OAK supported cameras...

Thanks @MartyG-RealSense for the support above

[MartyG-RealSense]

You are very welcome. I'm sincerely sorry that you did not get the outcome that you were aiming for.