1. Introduction to streaming
Push streaming is to push the input video data to the streaming media server. The input video data can be local video files (avi, mp4, flv...), memory video data, or system devices such as cameras, or is the network stream URL. This article introduces the method of pushing local video files to the RTMP streaming media server in the form of RTMP live streaming through FFmpeg programming.
The network topology structure of streaming is as follows:
RTMP streaming server: implemented by nginx+rtmp module
RTMP streamer:
RTMP streamer: implemented with ffmpeg
It should be noted that the encapsulation format used by RTMP is FLV. When specifying the output streaming media format, you need to specify its encapsulation format as "flv". Similarly, other streaming media protocols also need to specify their encapsulation formats. For example, when using UDP to push streaming media, you can specify its encapsulation format as "mpegts".
2. FFmpeg streaming
FFMpeg handles RTMP streams in two ways:
One is to use the built-in RTMP code function;
One is to use the third-party library librtmp;
These two methods are somewhat different
1. The RTMP code function that comes with FFmpeg
The RTMP code that comes with FFmpeg only supports the RTMP protocol, not the rtmpt, rtmpe, rtmpte and rtmps protocols;
The command line settings are as follows:
\1. Save the RTMP stream as it is in a file
# ./ffmpeg -i rtmp://192.168.1.11:1935/live/teststream -acodec copy -vcodec copy -f flv -y test.flv
\2. Save the RTMP stream transcoding as a file
# ./ffmpeg -i rtmp://192.168.1.11:1935/live/teststream -acodec ... -vcodec ... -f mp4 -y test.mp4
\3. Transcode the RTMP stream and then push it to the RTMP stream server in the form of RTMP stream
# ./ffmpeg -i rtmp://192.168.1.11:1935/live/teststream -acodec ... -vcodec ... -f flv rtmp://10.2.11.111/live/newstream
2. The third-party library librtmp
How to get FFMpeg to link this library (updated later)
FFMpeg can support rtmp://, rtmpt://, rtmpe://, rtmpte://, and rtmps:// protocols.
FFMpeg linked with librtmp accepts a string as input,
如："rtmp://server:port/app/playpath/stream_name live=1 playpath=xxx ..."
NOTE: Quotes are required;
\1. Save the RTMP live stream as it is and save it as a file:
# ./ffmpeg -i "rtmp://http://pub1.guoshi.com/live/newcetv1 live=1" -vcodec copy -acodec copy -y cetv1.flv
\2. Transcode the RTMP stream and then push it to the RTMP stream server in the form of RTMP stream
# ./ffmpeg -i "rtmp://192.168.1.11:1935/live/app/teststream live=1" -acodec ... -vcodec ... -f flv rtmp://10.2.11.111/live/newstream
\3. Play RTMP live stream with ffplay:
ffplay "rtmp://http://pub1.guoshi.com/live/newcetv1 live=1"
\4. When using the FFMPEG class library for programming, it is the same,
Just pass the string to avformat_open_input(), like:
ffplay "rtmp://http://pub1.guoshi.com/live/newcetv1 live=1"
char url[]="rtmp://http://live.hkstv.hk.lxdns.com/live/hks live=1";
avformat_open_input(&pFormatCtx,url,NULL,&avdic)
3. Flowchart of flow booster function
4. Code
int main(int argc, char * argv[])
{
AVFormatContext *pInFmtContext = NULL;
AVStream *in_stream;
AVCodecContext *pInCodecCtx;
AVCodec *pInCodec;
AVPacket *in_packet;
AVFormatContext * pOutFmtContext;
AVOutputFormat *outputFmt;
AVStream * out_stream;
//AVCodecContext * pOutCodecCtx;
//AVCodec *pOutCodec;
//AVPacket *out_packet;
//AVFrame *pOutFrame;
AVRational frame_rate;
double duration;
//int picture_size = 0;
//FILE *fp;
int ret;
const char * default_url = "rtmp://localhost:1935/live/tuiliu1";
char in_file[128] = {0};
char out_file[256] = {0};
int videoindex = -1;
int audioindex = -1;
int video_frame_count = 0;
int audio_frame_count = 0;
int video_frame_size = 0;
int audio_frame_size = 0;
int i;
int got_picture;
if(argc < 2){
printf("Usage: a.out <in_filename> <url>\n");
return -1;
}
memcpy(in_file, argv[1], strlen(argv[1]));
if( argc == 2){
memcpy(out_file, default_url, strlen(default_url));
}else{
memcpy(out_file, argv[2], strlen(argv[2]));
}
//av_register_all();
//avformat_network_init();
// Open an input stream and read the header,
if (avformat_open_input ( &pInFmtContext, in_file, NULL, NULL) < 0){
printf("avformat_open_input failed\n");
return -1;
}
//查询输入流中的所有流信息
if( avformat_find_stream_info(pInFmtContext, NULL) < 0){
printf("avformat_find_stream_info failed\n");
return -1;
}
av_dump_format(pInFmtContext, 0, in_file, 0);
ret = avformat_alloc_output_context2(&pOutFmtContext, NULL, "flv", out_file);
if(ret < 0){
printf("avformat_alloc_output_context2 failed\n");
return -1;
}
//outputFmt = pOutFmtContext->oformat;
for(i=0; i < pInFmtContext->nb_streams; i++){
in_stream = pInFmtContext->streams[i];
if( in_stream->codecpar->codec_type == AVMEDIA_TYPE_AUDIO){
audioindex = i;
}
if( in_stream->codecpar->codec_type == AVMEDIA_TYPE_VIDEO){
videoindex = i;
frame_rate = av_guess_frame_rate(pInFmtContext, in_stream, NULL);
printf("video: frame_rate:%d/%d\n", frame_rate.num, frame_rate.den);
printf("video: frame_rate:%d/%d\n", frame_rate.den, frame_rate.num);
duration = av_q2d((AVRational){frame_rate.den, frame_rate.num});
}
pInCodec = avcodec_find_decoder(in_stream->codecpar->codec_id);
printf("%x, %d\n", pInCodec, in_stream->codecpar->codec_id);
//printf("-----%s,%s\n", pInCodec->name, in_stream->codec->codec->name);
out_stream = avformat_new_stream(pOutFmtContext, pInCodec);//in_stream->codec->codec);
if( out_stream == NULL){
printf("avformat_new_stream failed:%d\n",i);
}
ret = avcodec_parameters_copy(out_stream->codecpar, in_stream->codecpar);
if( ret < 0){
printf("avcodec_parameters_copy failed:%d\n", i);
}
out_stream->codecpar->codec_tag = 0;
if( pOutFmtContext->oformat->flags & AVFMT_GLOBALHEADER){//AVFMT_GLOBALHEADER代表封装格式包含“全局头”（即整个文件的文件头），大部分封装格式是这样的。一些封装格式没有“全局头”，比如MPEG2TS
out_stream->codec->flags |= AV_CODEC_FLAG_GLOBAL_HEADER;
}
}
av_dump_format(pOutFmtContext, 0, out_file, 1);
ret = avio_open(&pOutFmtContext->pb, out_file, AVIO_FLAG_WRITE);
if(ret < 0){
printf("avio_open failed:%d\n", ret);
return -1;
}
int64_t start_time = av_gettime();
ret = avformat_write_header(pOutFmtContext, NULL);
in_packet = av_packet_alloc();
while(1){
ret = av_read_frame(pInFmtContext, in_packet);
if(ret < 0){
printf("read frame end\n");
break;
}
in_stream = pInFmtContext->streams[in_packet->stream_index];
if(in_packet->stream_index == videoindex){
video_frame_size += in_packet->size;
printf("recv %5d video frame %5d-%5d\n", ++video_frame_count, in_packet->size, video_frame_size);
}
if(in_packet->stream_index == audioindex){
audio_frame_size += in_packet->size;
printf("recv %5d audio frame %5d-%5d\n", ++audio_frame_count, in_packet->size, audio_frame_size);
}
int codec_type = in_stream->codecpar->codec_type;
if( codec_type == AVMEDIA_TYPE_VIDEO){
#if 0
//延时方案1: 根据 1/帧率 来计算延时时间
av_usleep((int64_t)(duration * AV_TIME_BASE));
//av_usleep(10);
printf("%d\n", (int)(duration * AV_TIME_BASE));
#else
// 延时方案2: 根据pts时间与系统时间的关系来计算延时时间， 该方案更优
AVRational dst_time_base = {1, AV_TIME_BASE};
int64_t pts_time = av_rescale_q(in_packet->pts, in_stream->time_base, dst_time_base);
int64_t now_time = av_gettime() - start_time;
if( pts_time > now_time)
av_usleep(pts_time - now_time);
//printf("%d\n", pts_time - now_time);
#endif
}
out_stream = pOutFmtContext->streams[in_packet->stream_index];
av_packet_rescale_ts(in_packet,in_stream->time_base, out_stream->time_base);
in_packet->pos = -1;
ret = av_interleaved_write_frame(pOutFmtContext, in_packet);
if( ret < 0){
printf("av_interleaved_write_frame failed\n");
break;
}
av_packet_unref(in_packet);
}
//
av_write_trailer(pOutFmtContext);
av_packet_free(&in_packet);
avformat_close_input(&pInFmtContext);
avio_close( pOutFmtContext->pb);
avformat_free_context(pOutFmtContext);
return 0;
}
有两点需要注意的地方
\1. 推流的速度
不能一下子将数据全推到服务器，这样流媒体服务器承受不住，实际中音频流的数据量相比视频要小很多，可以不必管它， 只按视频播放速度(帧率)来推流即可满足需要。因此每推送一个视频帧，要延时一个视频帧的时长。视频帧的时长可根据帧率计算得出，即 1/帧率。
上述代码中采用的是av_usleep()直接延时等待的方式, ， 等待时间为‘1/帧率’， 由于存在程序处理的时间，系统延时等， 这种方式控制时间是不准确的，但是上述代码却很直观的表现了推流延时的实现。 实际中，我们需要考虑系统执行的时间以及延时， 可以结合系统当前时间与视频帧pts时间之间的差距，来决定延时的时间，这样计算的延时时间相对更准确, 计算公式如下：
delay_time = pts_time - now_time
= av_rescale_q(pkt.dts, ifmt_ctx->streams[videoindex]->time_base, (AVRational){1,AV_TIME_BASE}) - (av_gettime() - start_time)
\2. 推流的类型
上述代码采用的推流协议是rtmp, rtmp推流必须推送flv封装格式，而其他的协议也有相应的格式要求(udp推流必须推送mpegts封装格式)。 如果要将上述代码修改为适配多种推流协议，则可根据推流协议自动选择相应的封装格式。
编译
gcc tuiliu1.c -lavformat -lavcodec -lavutil
验证
要验证推流程序是否正确，我们需要搭建一个nginx+rtmp流媒体服务器(搭建nginx+rtmp服务器)，而拉流端可以使用ffplay，参考以下过程：
1. 启动nginx服务器
nginx
2. 启动拉流
ffplay rtmp://localhost:1935/live/tuiliu1
3. 启动推流：
./a.out test.flv
接下来，就能看视频了
遗留问题
\1. 无论使用ffplay命令播放视频还是使用SDL编程播放视频，都会导致compiz占用cpu过高
也尝试过网上的多种解决方式，均无法解决，怀疑是compiz的bug
\2. 无论使用ffmpeg命令推流还是使用以上代码推流，都会在推流结束调用av_write_trailer时打印
[flv @ 0x858c440] Failed to update header with correct duration.
[flv @ 0x858c440] Failed to update header with correct filesize.
使用以下推流命令依然会打印以上信息。
ffmpeg -re -i test.mp4 -c copy -f flv rtmp://localhost:1935/live/tuiliu1
原文https://zhuanlan.zhihu.com/p/436334751
★文末名片可以免费领取音视频开发学习资料，内容包括（FFmpeg ，webRTC ，rtmp ，hls ，rtsp ，ffplay ，srs）以及音视频学习路线图等等。
见下方!↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓↓