覆盆子 PI 实时视频流

Question

我正在为树莓派相机使用各种视频流选项。目前具有最低延迟的最佳解决方案是将数据从raspividto管道传输nc到mplayer

在覆盆子上：

/opt/vc/bin/raspivid -t 0 -hf -vf -w 640 -h 480 --nopreview -o - | nc -l 5000

在客户端机器上（-fps 60跳过缓冲的技巧）

nc $RASP_IP 5000 | mplayer -nosound -framedrop -x 640 -y 480 -fps 60 -demuxer +h264es -cache 1024 -

这项工作非常好，几乎为零延迟。

现在我想将一些动态数据叠加到视频上。实现这一目标的最佳方法是什么？

我见过raspivid直接编辑和添加 opencv 之类的解决方案，但这在我的情况下不起作用，因为显示器必须位于与连接到相机的机器不同的机器上。

技术（语言/库）并不重要，除非它可以在 *nix 上运行（.NET 不是一个选项）

Answer 1

以下是我使用多播通过网络流式传输 Pi 摄像头的几个命令，并在图像底部写入时间和摄像头名称。我在每一行之间添加了注释来解释它的作用，但命令本身必须全部位于同一行，没有任何注释才能实际工作（一行以 开头v4l2-ctl，另一行以 开头ffmpeg）。

# Set up the camera device for dynamic framerate (to get better images in
# low light) and for uncompressed video output.
v4l2-ctl
  -d /dev/video0
  --set-ctrl=exposure_dynamic_framerate=1
  --set-ctrl=scene_mode=8
  -v width=1296,height=960,pixelformat=YU12

# Run ffmpeg.
ffmpeg

  # Select the frame size.  This is for the Pi camera v1 and must match the
  # size set above in v4l2-ctl.
  -f rawvideo -pix_fmt yuv420p -video_size 1296x960

  # Use the system clock because the camera stream doesn't have timestamps.
  -use_wallclock_as_timestamps 1

  # Select the first camera.  This could change if you have multiple
  # cameras connected (e.g. via USB)
  -i /dev/video0

  # Set the framerate in the output H264 stream.  This value is double
  # the actual framerate, so 30 means 15 fps.
  -bsf h264_metadata=tick_rate=30

  # Use a video filter
  -vf '

    # Make the video higher by 37 pixels (32 for the words and 5 for padding)
    pad=h=(in_h+5+32),

    # Write the time at the bottom right.
    drawtext=x=(w-tw-8):y=(h-28):fontcolor=white:fontsize=28:text=%{localtime},

    # Write the hostname on the bottom left, which we put in the file
    # /tmp/cam_hostname before running this command.
    drawtext=x=8:y=(h-28):fontcolor=white:fontsize=28:textfile=/tmp/cam_hostname,

    # Write another message from /tmp/cam_msg after the hostname.  This
    # file is read each frame, so it can contain live data such as the
    # temperature or motion sensor status.  The file contents must be
    # written by another program.
    drawtext=x=32:y=(h-28):fontcolor=white:fontsize=28:textfile=/tmp/cam_msg:reload=1

  # End of -vf parameter
  '

  # Duplicate or drop frames as needed to ensure a constant output framerate.
  # This is because the Pi camera is set to a dynamic framerate, so at night
  # the framerate will drop to give a brighter image.  This option ensures
  # that whatever the camera framerate, the output H264 framerate will be
  # constant.
  -vsync 1

  # Specify the output framerate.  We use 15 fps because that's the camera's
  # maximum in high resolution mode.
  -r 15

  # Use the hardware H264 encoder via the V4L2 M2M interface.
  -c:v h264_v4l2m2m

  # Set the bitrate of the output video.
  -b:v 5M

  # No audio.
  -an

  # Adjust the output H264 bitstream so the timestamps run at the correct
  # rate.  This is required so if the stream is recorded, media players will
  # seek accurately (e.g. seeking 10 seconds forward will go forward 10
  # seconds).  Without this, seeking forward by 10 seconds could jump forward
  # by a minute or more, making it difficult to seek around in the
  # recorded footage.
  -bsf 'setts=ts=N*(1/15)*100000,h264_metadata=tick_rate=30'

  # Output the final stream via RTP to a multicast IP address.
  -f rtp_mpegts udp://239.0.0.3:5004