如何使用 FFMPEG 驱动程序使 opencv 工作

我的 linuxbox 上有一个摄像头，它运行良好：

# $ ls -al /dev/video*
# crw-rw----+ 1 root video 81, 0 janv.  8 16:13 /dev/video0
# crw-rw----+ 1 root video 81, 1 janv.  8 16:13 /dev/video1
# $ groups
# adm cdrom sudo dip video plugdev lpadmin lxd sambashare docker libvirt

来自带有 cv2 的 python，它可以与默认驱动程序很好地配合CAP_V4L2

>>> from pathlib import Path
>>> import cv2
>>> print(cv2.VideoCapture(0, apiPreference=cv2.cv2.CAP_V4L2).isOpened())
True
>>>

我想用 FFMPEG 驱动程序访问它（没有成功）：

>>> print(cv2.VideoCapture(0, apiPreference=cv2.CAP_FFMPEG).isOpened())
False
>>>

从Python方面来看，opencv看起来有FFMPEG驱动程序：

  >>> cv2.__version__
  '4.4.0'
  >>> info = cv2.getBuildInformation()
  >>> video, parallel = info.index('Video'), info.index('Parallel')
  >>> print(info[video:parallel])
  Video I/O:
       DC1394:                      NO
       FFMPEG:                      YES
         avcodec:                   YES (58.109.100)
         avformat:                  YES (58.61.100)
         avutil:                    YES (56.60.100)
         swscale:                   YES (5.8.100)
         avresample:                NO
       GStreamer:                   NO
       v4l/v4l2:                    YES (linux/videodev2.h)
  >>>

从 Linux 方面看也不错：

$ dpkg -l |grep -i opencv
ii  libopencv-core4.2:amd64                    4.2.0+dfsg-5                          amd64        computer vision core library
ii  libopencv-imgcodecs4.2:amd64               4.2.0+dfsg-5                          amd64        computer vision Image Codecs library
ii  libopencv-imgproc4.2:amd64                 4.2.0+dfsg-5                          amd64        computer vision Image Processing library
ii  libopencv-videoio4.2:amd64                 4.2.0+dfsg-5                          amd64        computer vision Video I/O library

$ dpkg -l |grep -i ffm
ii  ffmpeg                                     7:4.2.4-1ubuntu0.1                    amd64        Tools for transcoding, streaming and playing of multimedia files
ii  gstreamer1.0-libav:amd64                   1.16.2-2                              amd64        ffmpeg plugin for GStreamer
ii  libavcodec-extra:amd64                     7:4.2.4-1ubuntu0.1                    amd64        FFmpeg library with extra codecs (metapackage)
ii  libavcodec-extra58:amd64                   7:4.2.4-1ubuntu0.1                    amd64        FFmpeg library with additional de/encoders for audio/video codecs
ii  libavdevice58:amd64                        7:4.2.4-1ubuntu0.1                    amd64        FFmpeg library for handling input and output devices - runtime files
ii  libavfilter7:amd64                         7:4.2.4-1ubuntu0.1                    amd64        FFmpeg library containing media filters - runtime files
ii  libavformat58:amd64                        7:4.2.4-1ubuntu0.1                    amd64        FFmpeg library with (de)muxers for multimedia containers - runtime files
ii  libavresample4:amd64                       7:4.2.4-1ubuntu0.1                    amd64        FFmpeg compatibility library for resampling - runtime files
ii  libavutil56:amd64                          7:4.2.4-1ubuntu0.1                    amd64        FFmpeg library with functions for simplifying programming - runtime files
ii  libffmpegthumbnailer4v5                    2.1.1-0.2build2                       amd64        shared library for ffmpegthumbnailer
ii  libpostproc55:amd64                        7:4.2.4-1ubuntu0.1                    amd64        FFmpeg library for post processing - runtime files
ii  libswresample3:amd64                       7:4.2.4-1ubuntu0.1                    amd64        FFmpeg library for audio resampling, rematrixing etc. - runtime files
ii  libswscale5:amd64                          7:4.2.4-1ubuntu0.1                    amd64        FFmpeg library for image scaling and various conversions - runtime files
$

OpenCV 是否使用 FFMPEG 从相机进行捕获？我认为它适用于文件（或流），因为 VideoCapture 既用于相机又用于文件/流。

在我的 Windows 设置中，用作从相机捕获的接口的 FFMPEG 也不起作用（但例如 CAP_DSHOW 有效），而作为文件的编解码器，它会读取帧：

cap = cv2.VideoCapture("..\\ooo.mp4", apiPreference=cv2.CAP_FFMPEG) # OK


>>> print(cv2.VideoCapture(0, apiPreference=cv2.CAP_FFMPEG).isOpened())
False
>>> print(cv2.VideoCapture(0, apiPreference=cv2.CAP_DSHOW).isOpened())
True


import cv2
# cap = cv2.VideoCapture(0, apiPreference=cv2.CAP_FFMPEG) # prints "ERROR"
cap = cv2.VideoCapture(0, apiPreference=cv2.CAP_DSHOW)  # OK, shows an image
ret, fr = cap.read()
if ret: cv2.imshow("FR", fr)
else: print("ERROR")
cv2.waitKey(0)

也可以看看：https://trac.ffmpeg.org/wiki/Capture/Webcam https://trac.ffmpeg.org/wiki/Capture/Webcam

"Linux

使用 video4linux2（或简称 v4l2）输入设备捕获实时视频 输入，例如来自网络摄像头的输入。请参阅 v4l2 输入设备文档 了解更多信息。 ”

对于 Windows，FFMPEG 被提到使用 DirectShow，如我的测试所示：

>Windows
>dshow
>
>Uses the dshow (DirectShow) input device which is the preferred option for
>Windows users.