V4l2 programming - bellbind/node-v4l2camera GitHub Wiki

About V4L2

Video4Linux2 (V4L2) is a Linux kernel API for streaming video/audio devices.

For accessing devices, program codes only use ioctl() and mmap() which params are the structs and the constants defined in linux/videodev2.h (and included headers).

For example, to get a captured frame is like:

-open device files

setup device with ioctl
prepare mmaped areas on deriver for captured frames with ioctl
share mmaped areas by querying with ioctl
poll (or select) device to capture
query the current frame mmaped area with ioctl
copy the frame pixel data from mmaped area to user memories

About UVC

UVC is common class of USB Video. e.g.

USB Webcam
Face cam upon the laptop PC display.

The UVC driver supports YUYV formats for frame pixels at least.

About YUYV pixel format

YUYV represents each horizontal two pixels to 4-byte as [Y0, U, Y1, V, ...].

The left side pixel RGB calculated by Y0, U, V
The right side pixel RGB calculated by Y1, U, V

Note: Pixel formats are not included info of width and height. For building bitmap from a frame, it also required width and height information.

YUYV is known as YUV422.

YUV is a color model used analog video devices and cables known as YCbCr.

Y represents luminance value of the picture. Only using Y values become a monochrome picture.

U and V represents 2-dimensional colors (blue and red). [U, V] becomes as:

Basic `ioctl()` flow for capturing

[1s] open a device file
[2s] check the device is capturing type
- VIDIOC_QUERYCAP
- VIDIOC_CROPCAP, VIDIOC_S_CAP
[2s] set capture settings
- VIDIOC_S_FMT: format and width/height
- VIDIOC_S_PARM: capture interval
[2s] prepare buffer on mmap
- VIDIOC_REQBUF (count > 2): prepare memories of frame count that required more than 2
- loop of prepared count
  - VIDIOC_QUERYBUF: acquire info for mmap of each frame
  - mmap: share captured buffer with device and user program
[3s] start capturing
- loop for mmap list
  - VIDIOC_QBUF: enqueue a mmap frame memory, then the memory is used for storing captured frame pixels
- VIDIOC_STREAMON: start capturing to enqueued memories
[4] loop of capturing each frame
- get captured frame pixel data
  - poll or select the device
  - VIDIOC_DQBUF: dequeue current captured memory and acquire index of mmap list
  - copy the mmap memory to user memory
  - VIDIOC_QBUF: enqueue dequeued mmap memory
[3e] stop capturing
- VIDIOC_STREAMOFF: stop capturing
[2e] release buffers
- munmap: release mmap memories
- VIDIOC_REQBUF (count = 0): release device memory by preparing memories of frame count as "0" (IMPORTANT)
[1e] close the device

for looping this process as: