Skip to main content

Multi‑Camera Capture Demo

This repository shows a minimal Python + OpenCV example that grabs still frames from up to eight USB webcams on Linux. Cameras are opened in pairs (to stay within the USB bus bandwidth limit), a frame is displayed in a 2 × 3 grid, and the handles are released before the next pair is opened.

Requirements​

componenttested version
Python≥ 3.9
OpenCV4.9+ (built with V4L2 support)
NumPy≥ 1.21
Matplotlib≥ 3.6
Linuxkernel 5.* / 6.* with uvcvideo

Install with mamba env create -f environment.yml or manually:

mamba create -n webcam python=3.11 opencv matplotlib numpy
mamba activate webcam

Quick start​

python getimage15_success.py        # grabs /dev/video1 … /dev/video8

All open‐error messages ("cam X: open failed", "read failed") are printed to STDOUT.

Why MJPEG?​

The problem: “No space left on device”​

When many USB‑2 webcams stream raw YUYV 4:2:2 frames, the kernel driver must allocate a large number of in‑flight buffers. Each 640 × 480 frame is 614 kB; at 30 fps even two cameras can exceed the default usbfs_memory_mb limit (16 MB). The V4L2 subsystem then returns -ENOSPC, exposed by FFmpeg / OpenCV as:

[video4linux2,v4l2] ioctl(VIDIOC_STREAMON): No space left on device

The solution: compressed MJPEG​

Most UVC cameras can encode Motion‑JPEG in hardware. The bulk data crossing the USB bus – and the temporary kernel buffers – shrink by a factor of 4‑10. For the same two 640 × 480 streams at 80 KiB each, you now need < 5 MB of kernel memory, well below the limit.

“Force MJPEG and it works fine – RAW eats buffer space until ENOSPC.” — ⁠opencv/opencv #23787 citeturn0search0

How to switch to MJPEG​

  1. Programmatically (preferred):

    cap = cv2.VideoCapture('/dev/video3')
    cap.set(cv2.CAP_PROP_FOURCC, cv2.VideoWriter_fourcc(*'MJPG'))

  2. Persistently with v4l2-ctl:

    sudo v4l2-ctl -d /dev/video3 \
         --set-fmt-video=width=640,height=480,pixelformat=MJPG

  3. Kernel quirk (if the camera ignores MJPG requests):

    sudo modprobe -r uvcvideo
    sudo modprobe uvcvideo quirks=128   # FIX_BANDWIDTH_QUirk

Script overview (getimage15_success.py)​

  • cam_ids = (2,4,6,8,10,12) – list of device numbers (IMPORTANT This may be different on your system - here it was multiple of two for some reason, 0 was inbuilt webcam)
  • Loop in steps of 2: open two cameras → grab one frame each → display in grid → release both.
  • GUI built with Matplotlib (plt.subplots(2, 3)).
  • Uses np.zeros placeholders for empty slots if you have < 6 cameras.
  • Press Esc to exit.

Troubleshooting​

symptomcausefix
open faileddevice busy (another process)fuser -v /dev/video* then kill
read failed after a whilecable power lossuse powered hub / USB‑3 port
green/pink tintcamera still in YUYVforce MJPG as above
high CPU usageMatplotlib renderingcall plt.pause(0.001) instead of tight sleep

License​

MIT

Acknowledgements​