Skip to content

Latest commit

 

History

History
113 lines (87 loc) · 3.82 KB

README.md

File metadata and controls

113 lines (87 loc) · 3.82 KB
Raw

Live HLS streaming with the Raspberry Pi camera.

I wanted to use the Raspberry Pi camera to encode video for HLS streaming, the format used by iPhones, iPads and many other kinds of device. HLS uses h264 encoding for video and AAC for audio.

Unfortunately I couldn't easily discover how to modify raspivid to produce videos in the precise format required for HLS. I expect that it's possible and that I didn't look hard enough.

The problem is this: the h264 bitstream consists of a series of NAL (Network Abstraction Layer) units. Each unit has a type (a unit is similar to a packet: just a chunk of data with a header); many contain compressed video data but two, SPS and PPS contain parameters about the dimensions of the video (amongst other things) that are needed to decode it properly.

Apple's HLS HTTP streaming protocol transfers chunks of video and/or audio in chunks typically between four and twelve seconds long. Each one of these chunks has to be a self contained video which means that it has to contain SPS and PPS NALs.

By default it seems that raspivid only places SPS and PPS at the start of the stream. That means that when the stream is chopped up into chunks for HLS transfer only the first of them - which contains the SPS and PPS is playable.

psips

The problem can be fixed by passing the h264 bit stream through a filter that

  • stores any SPS and PPS NALs it sees in the stream
  • places a copy of the most recent SPS and PPS just before any IDR (key) frames.

The reason for that is that we're next going to pass the stream to ffmpeg to be split into chunks a few seconds long and it can only split the stream at a key frame. Placing SPS and PPS right before the IDR frame means that ffmpeg will split the stream just before the SPS, PPS pair - which is exactly what we want.

To build it on your Raspberry Pi:

$ sudo apt-get install git build-essential autoconf libtool
$ cd /to/some/work/dir
$ git clone git://github.com/AndyA/psips.git
$ cd psips
$ ./setup.sh && ./configure && make && make install

You use it like this:

$ raspivid -w 1280 -h 720 -fps 25 -hf -t 86400000 -b 1800000 -o - | psips > live.h264

Or you could use it to add SPS and PPS for an existing file like this:

$ psips < oldfile.h264 > newfile.h264

and it works...

In the example directory there's a shell script, /examples/hls.sh, that packages HLS and makes it available for streaming via your webserver.

#!/bin/bash

base="/usr/share/nginx/www"

set -x

rm -rf live live.h264 "$base/live"
mkdir -p live
ln -s "$PWD/live" "$base/live"

# fifos seem to work more reliably than pipes - and the fact that the
# fifo can be named helps ffmpeg guess the format correctly.
mkfifo live.h264
raspivid -w 1280 -h 720 -fps 25 -hf -t 86400000 -b 1800000 -o - | psips > live.h264 &

# Letting the buffer fill a little seems to help ffmpeg to id the stream
sleep 2

# Need ffmpeg around 1.0.5 or later. The stock Debian ffmpeg won't work.
# I'm not aware of options apart from building it from source. I have
# Raspbian packags built from Debian Multimedia sources. Available on
# request but I don't want to post them publicly because I haven't cross
# compiled all of Debian Multimedia and conflicts can occur.
ffmpeg -y \
  -i live.h264 \
  -f s16le -i /dev/zero -r:a 48000 -ac 2 \
  -c:v copy \
  -c:a libfaac -b:a 128k \
  -map 0:0 -map 1:0 \
  -f segment \
  -segment_time 8 \
  -segment_format mpegts \
  -segment_list "$base/live.m3u8" \
  -segment_list_size 720 \
  -segment_list_flags live \
  -segment_list_type m3u8 \
  "live/%08d.ts" < /dev/null

At the time of writing that's been successfully streaming live video from my Raspberry Pi camera for quite a few hours. And the quality is astonishing for such a small camera. I'll link to some once it's daylight again and I have something worth sharing.

Andy Armstrong, [email protected]