This document contains collected notes regarding html5 live streaming approaches. I'm trying to understand how to build a system that enables streaming of live video to HTML5 clients. Following content is mainly centered around MPEG-DASH - modern way of dealing with given needs.
To prepare movie.avi from MPEG-DASH streaming we will execute following ffmpeg commands:
> ffmpeg -y -i movie.avi -an -c:v libx264 -x264opts 'keyint=24:min-keyint=24:no-scenecut' -b:v 1500k -maxrate 1500k -bufsize 3000k -vf "scale=-1:720" movie-720.mp4
> ffmpeg -y -i movie.avi -an -c:v libx264 -x264opts 'keyint=24:min-keyint=24:no-scenecut' -b:v 800k -maxrate 800k -bufsize 1600k -vf "scale=-1:540" movie-540.mp4
> ffmpeg -y -i movie.avi -an -c:v libx264 -x264opts 'keyint=24:min-keyint=24:no-scenecut' -b:v 400k -maxrate 400k -bufsize 800k -vf "scale=-1:360" movie-360.mp4
> ffmpeg -y -i movie.avi -vn -c:a aac -b:a 128k movie.m4a
It will generate m4a audio file and 3 mp4 video files with different bitrate variants: 1500k, 800k and 400k. Now, it's time to fragment them into video segments and generate the MPEG-DASH Manifest file. We'll do it using mp4box utility:
> mp4box -dash-strict 2000 -rap -frag-rap -bs-switching no -profile "dashavc264:live" -out movie-dash.mpd movie-720.mp4 movie-720.mp4 movie-720.mp4 movie.m4a
Resources:
- https://blog.streamroot.io/encode-multi-bitrate-videos-mpeg-dash-mse-based-media-players-22/
- https://rybakov.com/blog/mpeg-dash/
It looks like there is a possibility to generate mpeg-dash content with single ffmpeg command, but I'm not sure how it works with multi-bitrate video variants:
> ffmpeg -y -re -i movie.mp4 ^
-c:v libx264 -x264opts "keyint=24:min-keyint=24:no-scenecut" -r 24 ^
-c:a aac -b:a 128k ^
-bf 1 -b_strategy 0 -sc_threshold 0 -pix_fmt yuv420p ^
-map 0:v:0 -map 0:a:0 -map 0:v:0 -map 0:a:0 -map 0:v:0 -map 0:a:0 ^
-b:v:0 250k -filter:v:0 "scale=-2:240" -profile:v:0 baseline ^
-b:v:1 750k -filter:v:1 "scale=-2:480" -profile:v:1 main ^
-b:v:2 1500k -filter:v:2 "scale=-2:720" -profile:v:2 high ^
-use_timeline 1 -use_template 1 -window_size 5 -adaptation_sets "id=0,streams=v id=1,streams=a" ^
-f dash movie-dash\movie.mpd
TODO: explain options ^^
The -re option tells ffmpeg to read the input file in realtime and not in the standard as-fast-as-possible manner.
Command above do all the needed conversion on the fly, but this could be optimized by dividing the process into two steps:
-
prepare output video with multiple video/audio streams which corresponds do adaptive DASH streams. Consider
-preset slow> ffmpeg -y -i sample.divx ^ -c:v libx264 -x264opts "keyint=24:min-keyint=24:no-scenecut" -r 24 ^ -c:a aac -b:a 128k ^ -bf 1 -b_strategy 0 -sc_threshold 0 -pix_fmt yuv420p ^ -map 0:v:0 -map 0:a:0 -map 0:v:0 -map 0:a:0 -map 0:v:0 -map 0:a:0 ^ -b:v:0 250k -filter:v:0 "scale=-2:240" -profile:v:0 baseline ^ -b:v:1 750k -filter:v:1 "scale=-2:480" -profile:v:1 main ^ -b:v:2 1500k -filter:v:2 "scale=-2:720" -profile:v:2 high ^ sample_dash.mp4 -
stream prepared output video with minimal dash configuration including
-reoption to enable pseudo-live streaming> mkdir sample_dash > cd sample_dash > ffmpeg -y -re -i ..\sample_dash.mp4 ^ -map 0 ^ -use_timeline 1 -use_template 1 -window_size 5 -adaptation_sets "id=0,streams=v id=1,streams=a" ^ -f dash sample.mpd
Resources:
- https://www.ffmpeg.org/ffmpeg-formats.html#dash-2
- https://bitmovin.com/video-bitrate-streaming-hls-dash/
- https://trac.ffmpeg.org/wiki/Map
- https://trac.ffmpeg.org/wiki/Encode/H.264#Profile
- https://sites.google.com/site/linuxencoding/x264-ffmpeg-mapping
- https://trac.ffmpeg.org/wiki/Scaling#KeepingtheAspectRatio
Hey du, thanks for sharing this content.
Just one question, why yout didn't specify the time of video segments ?