-
Notifications
You must be signed in to change notification settings - Fork 41
multithreading_opencv.py cannot handle 3 or more camera #161
Comments
What camera model are you using, especially what interface and how are they connected to the PC? |
all 4 cameras in my hand are 1800 U-500m (USB3, monochrome camera, resolution 2,592 x 1,944). All camera are connected to the pc with USB3 cable. All camera stream in Vimba Viewer. |
My questions were not worded clear enough: Do they all stream simulatenously in Vimba Viewer? |
something interesting happened. all camera can stream simultaneously in Vimba Viewer on Windows10 but not working on Ubuntu18. I will try to run my code on windows10 to see if it works. here are usb hubs: |
For Linux based systems we recommend setting a larger USBFS Buffer.
Syslinux
|
I am trying to set up a streaming system using three allied cameras (5k monochrome). I am currently using the multithreading_opencv.py script, but I've encountered an issue where only two threads are working after I disabled the set_nearest_value function in the setup_camera method to avoid cropping the image and ensure that the entire frame is displayed on the screen.while the others are not running. To identify the problem, I added a print(cam.get_id()) statement in the FrameProducer's call function. However, it only prints the IDs of at most two cameras, even though there are three cameras connected to the computer.
I noticed that only one CPU core is being utilized, and the memory usage remains around 7%. So I don't think there are band width issue.
I have core i7 9700k and 16gb ram. code are running on ubuntu 18
Could you please assist me in resolving these issues?
Here is the code:
The text was updated successfully, but these errors were encountered: