I, like many other brave tinkerers before me, thought it would be a simple task to take an old USB camera (c920) can pair it with a raspberry pi to make a network streaming device (e.g., baby monitor). As those that have gone before me, I have now realized (after two days of tearing my hair out), that this is an extremely complicated task.
Problem statement: I have a raspberry pi zero and a c920 webcam. I want to use the H.264 bitstream from the webcam and serve it on the pi without transcoding it (the feeble processor would really struggle). I want to combine the video stream with its audio and send it over to a browser (phone, tablet, pc - something HTML5 without NAPI).
My current strategy is to do the following:
ffmpeg -re -f s16le -i /dev/zero -f v4l2 -thread_queue_size 512 -codec:v h264 -s 1920x1080 -i /dev/video0 -codec:v copy -acodec aac -ab 128k -g 50 http://localhost:8090/camera.ffm (this is with dummy audio - I figured I would add audio later)
sudo ffserver -d -f /etc/ffserver.conf to received the feed and broadcast it as a stream. This is the
`HTTPPort 8090 HTTPBindAddress 0.0.0.0 MaxHTTPConnections 2000 MaxClients 1000 MaxBandwidth 100000 CustomLog - <Feed camera.ffm> File /tmp/streamwebm.ffm FileMaxSize 50M ACL allow localhost ACL allow 184.108.40.206 ACL allow 127.0.0.1 ACL allow 192.168.0.0 192.168.0.255 </Feed> <Stream stream> ACL allow 192.168.0.0 192.168.0.255 Format webm # Video Settings VideoFrameRate 30 VideoSize 1920x1080 # Audio settings AudioCodec libvorbis AudioSampleRate 48000 AVOptionAudio flags +global_header MaxTime 0 AVOptionVideo me_range 16 AVOptionVideo qdiff 4 AVOptionVideo qmin 4 AVOptionVideo qmax 40 #AVOptionVideo good AVOptionVideo flags +global_header # Streaming settings PreRoll 10 StartSendOnKey Metadata author "author" Metadata copyright "copyright" Metadata title "Web app name" Metadata comment "comment" </stream>
My basic html is
<html><head></head><body><video> <source src="http://localhost:8090/stream"> </video></body></html>
And the browser on the client says
Thoughts: - https://stackoverflow.com/questions/28435564/begin-stream-simple-mp4-with-ffserver explains that ffserver can't stream .mp4 because of headers or something. This is why I am using webm (which doesn't support h.264 I believe and is causing the really slow performance converting to vp9). I'm not concerned about CPU usage at the moment, just want to get an image to appear on the browser!
I hear one issue deals with 'chunking' - that the camera h.264 is a bitstream but h.264 streams for html5 should be chunked. Not sure how that would work.
I have tried VLC for some things (RTP) but haven't have success.
Most resources (SE and other sites) are from 2010-2015 and it seems as thought v4l2 and other things have developed since then.
As my problem is most likely general ignorance of the subject matter, I would appreciate any answers that provide some general understanding as to the theory behind different techniques. I know this makes the question more of a call for opinion and less appropriate for SE, but I'm fixing to throw my computer out the window (you know the feeling).