We've been doing some experimentation with ffmpeg to encode video for live streaming and for recording to files for on-demand playback.  While I've been impressed by the capabilities of ffmpeg, I've found that its command-line processing is quite daunting.  Here I provide a set of command-line options along with commentary on what is going on.

One of the most frustrating things I've found in trying to learn ffmpeg is that many examples online are quite out-of-date.  It seems that ffmpeg has changed its command-line options fairly frequently.  So many of the examples you'll find won't work.

The example that follows uses a snapshot of the ffmpeg subversion code from September 10, 2009.

Ready?

Are you sure?

Here is the command line:

ffmpeg -f rawvideo -pix_fmt yuv422p \
    -s 720x486 -r 29.97 \
    -i /tmp/vpipe \
    -ar 48000 -f s16le -ac 2 -i /tmp/apipe \
    -vol 4096 \
    -acodec libfaac -ac 2 -ab 192k -ar 44100 \
    -async 1 \
    -vcodec libx264 -vpre default -threads 0 \
    -croptop 6 -cropbottom 6 -cropleft 9 -cropright 9 \
    -r 20 -g 45 -s 432x324 -b 1024k -bt 100k \
    -deinterlace \
    -y \
    '/tmp/encoding-0001.mp4' \
    -an \
    -vcodec libx264 -vpre default -threads 0 \
    -croptop 6 -cropbottom 6 -cropleft 9 -cropright 9 \
    -r 20 -g 45 -s 432x324 -b 1024k -bt 100k \
    -deinterlace \
    -vglobal 1 \
    -f rtp rtp://127.0.0.1:9012 \
    -vn \
    -vol 4096 \
    -acodec libfaac -ac 2 -ab 192k -ar 44100 \
    -async 1 \
    -flags +global_header \
    -f rtp rtp://127.0.0.1:9014 \
    -newaudio \
    -an \
    -vcodec libx264 -vpre default -threads 0 \
    -croptop 6 -cropbottom 6 -cropleft 9 -cropright 9 \
    -r 15 -g 45 -s 432x324 -b 768k -bt 50k \
    -deinterlace \
    -vglobal 1 \
    -f rtp rtp://127.0.0.1:9008 \
    -newvideo \
    -vn \
    -vol 4096 \
    -acodec libfaac -ac 2 -ab 128k -ar 44100 \
    -async 1 \
    -flags +global_header \
    -f rtp rtp://127.0.0.1:9010 \
    -newaudio

Don't be intimidated.  We'll break this thing down line-by-line.

Specifying the input

ffmpeg -f rawvideo -pix_fmt yuv422p \
    -s 720x486 -r 29.97 \
    -i /tmp/vpipe \

Everything up to the -i option describes what is contained in the input file /tmp/vpipe.  In this case, our input video is raw (uncompressed) frame data in the YUV 4:2:2 planar format, 720 pixels wide by 486 pixels high, with a frame rate of 29.97 frames per second.  Note that these options must precede the -i option.  If any of these options came after the -i, ffmpeg would think that they belonged to the next input file specified.

    -ar 48000 -f s16le -ac 2 -i /tmp/apipe \

Now we're telling ffmpeg what the audio looks like in the input file /tmp/apipe.  We have a sample rate of 48000 samples per second; each sample is signed 16-bit, little endian, and there are 2 audio channels.

Specifying the output

The next set of options describes the output format, for both audio and video.  I do not believe that the order within these options is critical, but I like to group them logically so that all the audio options are together and all the video options are together.

    -vol 4096 \
    -acodec libfaac -ac 2 -ab 192k -ar 44100 \
    -async 1 \

The -vol option indicates that we're going to adjust the audio levels.  The baseline value is 256.  I have not seen good documentation on the valid values here, but I've used values like 512, 1024, 2048, and 4096 to bump up the volume.  Picking the right value here requires some experimentation and depends heavily on the levels in your original source.

We're going to output AAC audio using the libfaac encoder.  ffmpeg has a built-in AAC encoder, but it doesn't seem to be as robust as that provided by libfaac.   If you want AAC, and your version of ffmpeg was linked against libfaac, I would recommend that you use it.

The AAC will be 2 channels, 192kbps, with a sample rate of 44100 Hz.

Finally, the -async option indicates that ffmpeg should use audio/video sync method 1, which adjusts the start of the video and audio tracks, but does not do any time stretching over the course of the tracks.

    -vcodec libx264 -vpre default -threads 0 \
    -croptop 6 -cropbottom 6 -cropleft 9 -cropright 9 \
    -r 20 -g 45 -s 432x324 -b 1024k -bt 100k \
    -deinterlace \

Here we are specifying our video output options.  We're going to use H.264 compression, as provided by the libx264 library.  The -vpre option means that we will use the default quality preset (note that when using libx264 to encode, you can specify two -vpre options. The first is quality, and the second is the profile to use (main, baseline, etc., with "main" being the default). I have not seen this documented very well.

The -threads 0 option instructs ffmpeg to use the optimal number of threads when encoding.

We are going to crop our video a few pixels around the border, as we were getting some noise around the edges of our source video.

The -r option specifies that our output will be 20 frames per second.  The -g option is the "group of pictures" (GOP) size, which is the number of frames between keyframes.  With a smaller number, your output will have more keyframes, which means that streaming clients will be able to recover more quickly if they drop packets for some reason.  It also will have a detrimental effect on file size.

The -s option specifies our frame size, the -b option specifies the desired bitrate, and the -bt option is the bitrate tolerance.  ffmpeg will try to keep the video close to the desired bitrate, and the tolerance tells it how much leeway it has above and below the target bitrate.

Finally, we are deinterlacing the video, as the source was NTSC interlaced video.  Without deinterlacing, you can see some very unpleasant "comb" artifacts in your digitized video.

    -y \
    '/tmp/encoding-0001.mp4' \

Here we specify the output file.  The -y option instructs ffmpeg to overwrite the file without asking for confirmation.  ffmpeg will infer the file format from the filename extension.  Here it will be writing to an MPEG4 file.

Adding another output

Now we are going to add some specs for RTP output.  We will send this RTP stream over the network to a Wowza server, which can convert the RTP to RTMP for playback in Flash clients.

Unlike when we wrote to an MPEG4 file,  RTP requires that we break the audio and video into two separate streams.

    -an \
    -vcodec libx264 -vpre default -threads 0 \
    -croptop 6 -cropbottom 6 -cropleft 9 -cropright 9 \
    -r 20 -g 45 -s 432x324 -b 1024k -bt 100k \
    -deinterlace \
    -vglobal 1 \
    -f rtp rtp://127.0.0.1:9012 \

You'll notice that most of these specs are the same as what we specified for our MPEG4 output.  But there are a few differences.  Let's focus on those.  The -an option tells ffmpeg to remove the audio stream from this output.  The -vglobal 1 option instructs ffmpeg to use out-of-band global headers in the video stream (as opposed to in-band).  This may help some players interpret the video stream.

Finally, we specify the output format as "rtp" with the -f option, and instead of a filename, we have a URL that indicates where ffmpeg should send the RTP packets.

Next we specify the audio output:

    -vn \
    -vol 4096 \
    -acodec libfaac -ac 2 -ab 192k -ar 44100 \
    -async 1 \
    -flags +global_header \
    -f rtp rtp://127.0.0.1:9014 \
    -newaudio \

Again, this is very similar to the options we provided for our MPEG4 output, with some notable differences.  First, we use the -vn option to indicate that this output does not contain video.  The -flags +global_header is used to force ffmpeg to spit out some important audio specifications in the SDP it generates (you use an SDP file on the Wowza server to connect the RTMP stream to the RTP stream; Wowza needs to know all about the audio and video to interpret it properly).

Again, we specify the rtp format with the -f option, and we provide a URL.  Note that the port number is different.  It is customary for RTP streams to use two ports with an open port between them.  The ports just after each of the RTP ports will be used for RTCP ports (in our example, ports 9013 and 9015 wil be used), where the receiver communicates back to the sender.

The last option, -newaudio, restores the audio stream that got killed off by the -an option earlier.  Note that -newaudio is a special option; it only modifies the output which immediately precedes it.  This is one of those cases where order of options is critical.

With these options, we are now writing to an MPEG4 file and streaming RTP simultaneously.  But wait, there's more...

Adding another RTP stream

Our first RTP stream uses about 1200 Kbps with audio and video combined.  Let's create an option for our bandwidth-deprived visitors to use.

We can tack on another pair of outputs, one for video and one for audio:

    -an \
    -vcodec libx264 -vpre default -threads 0 \
    -croptop 6 -cropbottom 6 -cropleft 9 -cropright 9 \
    -r 15 -g 45 -s 432x324 -b 256k -bt 50k \
    -deinterlace \
    -vglobal 1 \
    -f rtp rtp://127.0.0.1:9008 \
    -newvideo \
    -vn \
    -vol 4096 \
    -acodec libfaac -ac 1 -ab 64k -ar 44100 \
    -async 1 \
    -flags +global_header \
    -f rtp rtp://127.0.0.1:9010 \
    -newaudio

There are a few differences between our second set of RTP outputs and the first:

  • the video framerate is 15 (-r 15)
  • the video bitrate is 256kbps and the tolerance is 50kbps (-b 256k -bt 50k)
  • the audio is single channel (-ac 1)
  • the audio is 64kbps (-ab 64k)
  • the ports are different in the RTP URLs

You'll also notice that there is a -newvideo option after the video RTP URL.  That's because our previous output was audio-only, due to the -vn option.  Using the -newvideo option restores the video stream to this output.  Without it, you'd have no audio (because of the -an), and no video (because of the -vn in the previous output).

So now we're writing to an MPEG4 file and streaming at two different bitrates simultaneously!  Pretty sweet.

I won't get into the specifics here of how you get the SDP file from ffmpeg, put it on the Wowza server, and connect to the Wowza server with a flash-based player.  There are lots of docs and notes in the Wowza forums on how to do that.

with ffmpeg to encode video for live streaming and for recording to files for on-demand playback的更多相关文章

  1. [quote ]ffmpeg, gstreamer, Raspberry Pi, Windows Desktop streaming

    [quote ]ffmpeg, gstreamer, Raspberry Pi, Windows Desktop streaming http://blog.pi3g.com/2013/08/ffmp ...

  2. ffmpeg的内部Video Buffer管理和传送机制

    ffmpeg的内部Video Buffer管理和传送机制 本文主要介绍ffmpeg解码器内部管理Video Buffer的原理和过程,ffmpeg的Videobuffer为内部管理,其流程大致为:注册 ...

  3. Video Codecs by FOURCC 视频格式编码

    FOURCC Name Summary 1978 A.M.Paredes predictor This is a LossLess video codec. >>> 2VUY 2VU ...

  4. Install FFmpeg, Mplayer, Mencoder, MP4Box, Flvtool2

    You can use the following tutorial to install ffmpeg and other video modules in your centos server.F ...

  5. Linux下编译带x264的ffmpeg的配置方法,包含SDL2

    一.环境准备 ffmpeg下载:http://www.ffmpeg.org/download.html x264下载:http://download.videolan.org/x264/snapsho ...

  6. 基于ffmpeg的简单音视频编解码的例子

    近日需要做一个视频转码服务器,对我这样一个在该领域的新手来说却是够我折腾一番,在别人的建议下开始研究开源ffmpeg项目,下面是在代码中看到的一 段例子代码,对我的学习非常有帮助.该例子代码包含音频的 ...

  7. 最简单的基于FFmpeg的编码器-纯净版(不包含libavformat)

    ===================================================== 最简单的基于FFmpeg的视频编码器文章列表: 最简单的基于FFMPEG的视频编码器(YUV ...

  8. Linux---centos编译安装ffmpeg

    环境 系统环境:CentOS release 6.7 (Final) 需求 编译安装ffmpeg 获取依赖 安装依赖包 yum install -y autoconf automake cmake f ...

  9. FFmpeg使用基础

    本文为作者原创,转载请注明出处:https://www.cnblogs.com/leisure_chn/p/10297002.html 本文介绍FFmpeg最基础的概念,了解FFmpeg的简单使用,帮 ...

随机推荐

  1. K贪心

    <span style="color:#330099;">/* K - 贪心 基础 Time Limit:1000MS Memory Limit:32768KB 64b ...

  2. 自增运算a++和++b(1)

    #include<reg52.h> #define uint unsigned int #define uchar unsigned char uchar code f[]={0x3f,0 ...

  3. 1. GDAL与OpenCV2.X数据转换(适合多光谱和高光谱等多通道的遥感影像)

    一.前言 GDAL具有强大的图像读写功能,但是对常用图像处理算法的集成较少,OpenCV恰恰具有较强的图像处理能力,因此有效的结合两者对图像(遥感影像)的处理带来了极大的方便.那么如何实现GDAL与o ...

  4. Java中String直接赋字符串和new String的区别

    解析Java中的String对象的数据类型 1. String是一个对象.  因为对象的默认值是null,所以String的默认值也是null:但它又是一种特殊的对象,有其它对象没有的一些特性. 2. ...

  5. .NET基础拾遗(1)类型语法基础和内存管理基础1

    一.基础类型和语法 1.1 .NET中所有类型的基类是什么? 在.NET中所有的内建类型都继承自System.Object类型. 1.2 值类型和引用类型的区别? 在.NET中的类型分为值类型和引用类 ...

  6. Cisco cmd 命令

    1.enable 开启全局配置模式:disable 禁用配置模式 2.config进入配置模式 3.line 设置进入用户模式密码:分为 line aux 0;line console 0;line ...

  7. OAuth协议与第三方登录:(QQ,百度,微信,微博)

    OAuth 相当于授权的U盾,提供第三方认证的协议,是个安全相关的协议,作用在于,使用户授权第三方的应用程序访问用户的web资源,并且不需要向第三方应用程序透露自己的密码. 传统互联网:应用于PC端, ...

  8. activiti笔记四 关于部署信息表act_re_deployment

    一.简要描述 部署流程定义时需要被持久化保存下来的信息.二.表结构说明 字段名称 字段描述 数据类型 主键 为空 取值说明 ID_ ID_ nvarchar(64) √ 主键ID NAME_ 部署名称 ...

  9. sqlCacheDependency 更新缓存Cache

    第一步 修改web,config <!--定义数据库连接--> <connectionStrings>  <add name="NorthwindConnect ...

  10. Apple Catching(POJ 2385)

    Apple Catching Time Limit: 1000MS   Memory Limit: 65536K Total Submissions: 9978   Accepted: 4839 De ...