Muxing Android MediaCodec encoded H264 packets into RTMP

Question

I am coming from a thread Encoding H.264 from camera with Android MediaCodec. My setup is very similar. However, I attempt to write mux the encoded frames and with javacv and broadcast them via rtmp.

RtmpClient.java

...
private volatile BlockingQueue<byte[]> mFrameQueue = new LinkedBlockingQueue<byte[]>(MAXIMUM_VIDEO_FRAME_BACKLOG);
...
private void startStream() throws FrameRecorder.Exception, IOException {
    if (TextUtils.isEmpty(mDestination)) {
        throw new IllegalStateException("Cannot start RtmpClient without destination");
    }

    if (mCamera == null) {
        throw new IllegalStateException("Cannot start RtmpClient without camera.");
    }

    Camera.Parameters cameraParams = mCamera.getParameters();

    mRecorder = new FFmpegFrameRecorder(
            mDestination,
            mVideoQuality.resX,
            mVideoQuality.resY,
            (mAudioQuality.channelType.equals(AudioQuality.CHANNEL_TYPE_STEREO) ? 2 : 1));

    mRecorder.setFormat("flv");

    mRecorder.setFrameRate(mVideoQuality.frameRate);
    mRecorder.setVideoBitrate(mVideoQuality.bitRate);
    mRecorder.setVideoCodec(avcodec.AV_CODEC_ID_H264);

    mRecorder.setSampleRate(mAudioQuality.samplingRate);
    mRecorder.setAudioBitrate(mAudioQuality.bitRate);
    mRecorder.setAudioCodec(avcodec.AV_CODEC_ID_AAC);

    mVideoStream = new VideoStream(mRecorder, mVideoQuality, mFrameQueue, mCamera);
    mAudioStream = new AudioStream(mRecorder, mAudioQuality);

    mRecorder.start();

    // Setup a bufferred preview callback
    setupCameraCallback(mCamera, mRtmpClient, DEFAULT_PREVIEW_CALLBACK_BUFFERS,
            mVideoQuality.resX * mVideoQuality.resY * ImageFormat.getBitsPerPixel(
                    cameraParams.getPreviewFormat())/8);

    try {
        mVideoStream.start();
        mAudioStream.start();
    }
    catch(Exception e) {
        e.printStackTrace();
        stopStream();
    }
}
...
@Override
public void onPreviewFrame(byte[] data, Camera camera) {
    boolean frameQueued = false;

    if (mRecorder == null || data == null) {
        return;
    }

    frameQueued = mFrameQueue.offer(data);

    // return the buffer to be reused - done in videostream
    //camera.addCallbackBuffer(data);
}
...

VideoStream.java

...
@Override
public void run() {
    try {
        mMediaCodec = MediaCodec.createEncoderByType("video/avc");
        MediaFormat mediaFormat = MediaFormat.createVideoFormat("video/avc", mVideoQuality.resX, mVideoQuality.resY);
        mediaFormat.setInteger(MediaFormat.KEY_BIT_RATE, mVideoQuality.bitRate);
        mediaFormat.setInteger(MediaFormat.KEY_FRAME_RATE, mVideoQuality.frameRate);
        mediaFormat.setInteger(MediaFormat.KEY_COLOR_FORMAT, MediaCodecInfo.CodecCapabilities.COLOR_FormatYUV420SemiPlanar);
        mediaFormat.setInteger(MediaFormat.KEY_I_FRAME_INTERVAL, 1);
        mMediaCodec.configure(mediaFormat, null, null, MediaCodec.CONFIGURE_FLAG_ENCODE);
        mMediaCodec.start();
    }
    catch(IOException e) {
        e.printStackTrace();
    }

    long startTimestamp = System.currentTimeMillis();
    long frameTimestamp = 0;
    byte[] rawFrame = null;

    try {
        while (!Thread.interrupted()) {
            rawFrame = mFrameQueue.take();

            frameTimestamp = 1000 * (System.currentTimeMillis() - startTimestamp);

            encodeFrame(rawFrame, frameTimestamp);

            // return the buffer to be reused
            mCamera.addCallbackBuffer(rawFrame);
        }
    }
    catch (InterruptedException ignore) {
        // ignore interrup while waiting
    }

    // Clean up video stream allocations
    try {
        mMediaCodec.stop();
        mMediaCodec.release();
        mOutputStream.flush();
        mOutputStream.close();
    } catch (Exception e){
        e.printStackTrace();
    }
}
...
private void encodeFrame(byte[] input, long timestamp) {
    try {
        ByteBuffer[] inputBuffers = mMediaCodec.getInputBuffers();
        ByteBuffer[] outputBuffers = mMediaCodec.getOutputBuffers();

        int inputBufferIndex = mMediaCodec.dequeueInputBuffer(0);

        if (inputBufferIndex >= 0) {
            ByteBuffer inputBuffer = inputBuffers[inputBufferIndex];
            inputBuffer.clear();
            inputBuffer.put(input);
            mMediaCodec.queueInputBuffer(inputBufferIndex, 0, input.length, timestamp, 0);
        }

        MediaCodec.BufferInfo bufferInfo = new MediaCodec.BufferInfo();

        int outputBufferIndex = mMediaCodec.dequeueOutputBuffer(bufferInfo, 0);

        if (outputBufferIndex >= 0) {
            while (outputBufferIndex >= 0) {
                ByteBuffer outputBuffer = outputBuffers[outputBufferIndex];

                // Should this be a direct byte buffer?
                byte[] outData = new byte[bufferInfo.size - bufferInfo.offset];
                outputBuffer.get(outData);

                mFrameRecorder.record(outData, bufferInfo.offset, outData.length, timestamp);

                mMediaCodec.releaseOutputBuffer(outputBufferIndex, false);
                outputBufferIndex = mMediaCodec.dequeueOutputBuffer(bufferInfo, 0);
            }
        }
        else if (outputBufferIndex == MediaCodec.INFO_OUTPUT_BUFFERS_CHANGED) {
            outputBuffers = mMediaCodec.getOutputBuffers();
        } else if (outputBufferIndex == MediaCodec.INFO_OUTPUT_FORMAT_CHANGED) {
            // ignore for now
        }
    } catch (Throwable t) {
        t.printStackTrace();
    }

}
...

FFmpegFrameRecorder.java

...
// Hackish codec copy frame recording function
public boolean record(byte[] encodedData, int offset, int length, long frameCount) throws Exception {
    int ret;

    if (encodedData == null) {
        return false;
    }

    av_init_packet(video_pkt);

    // this is why i wondered whether I should get outputbuffer data into direct byte buffer
    video_outbuf.put(encodedData, 0, encodedData.length);

    video_pkt.data(video_outbuf);
    video_pkt.size(video_outbuf_size);

    video_pkt.pts(frameCount);
    video_pkt.dts(frameCount);

    video_pkt.stream_index(video_st.index());

    synchronized (oc) {
        /* write the compressed frame in the media file */
        if (interleaved && audio_st != null) {
            if ((ret = av_interleaved_write_frame(oc, video_pkt)) < 0) {
                throw new Exception("av_interleaved_write_frame() error " + ret + " while writing interleaved video frame.");
            }
        } else {
            if ((ret = av_write_frame(oc, video_pkt)) < 0) {
                throw new Exception("av_write_frame() error " + ret + " while writing video frame.");
            }
        }
    }
    return (video_pkt.flags() & AV_PKT_FLAG_KEY) == 1;
}
...

When I try to stream the video and run ffprobe on it, I get the following output:

ffprobe version 2.5.3 Copyright (c) 2007-2015 the FFmpeg developers
  built on Jan 19 2015 12:56:57 with gcc 4.1.2 (GCC) 20080704 (Red Hat 4.1.2-55)
  configuration: --prefix=/usr --bindir=/usr/bin --datadir=/usr/share/ffmpeg --incdir=/usr/include/ffmpeg --libdir=/usr/lib64 --mandir=/usr/share/man --arch=x86_64 --optflags='-O2 -g -pipe -Wall -Wp,-D_FORTIFY_SOURCE=2 -fexceptions -fstack-protector --param=ssp-buffer-size=4 -m64 -mtune=generic' --enable-bzlib --disable-crystalhd --enable-libass --enable-libdc1394 --enable-libfaac --enable-nonfree --disable-indev=jack --enable-libfreetype --enable-libgsm --enable-libmp3lame --enable-openal --enable-libopencv --enable-libopenjpeg --enable-libopus --enable-librtmp --enable-libtheora --enable-libvorbis --enable-libvpx --enable-libx264 --enable-libxvid --enable-x11grab --enable-avfilter --enable-avresample --enable-postproc --enable-pthreads --disable-static --enable-shared --enable-gpl --disable-debug --disable-stripping --enable-libcaca --shlibdir=/usr/lib64 --enable-runtime-cpudetect
  libavutil      54. 15.100 / 54. 15.100
  libavcodec     56. 13.100 / 56. 13.100
  libavformat    56. 15.102 / 56. 15.102
  libavdevice    56.  3.100 / 56.  3.100
  libavfilter     5.  2.103 /  5.  2.103
  libavresample   2.  1.  0 /  2.  1.  0
  libswscale      3.  1.101 /  3.  1.101
  libswresample   1.  1.100 /  1.  1.100
  libpostproc    53.  3.100 / 53.  3.100
Metadata:
  Server                NGINX RTMP (github.com/arut/nginx-rtmp-module)
  width                 320.00
  height                240.00
  displayWidth          320.00
  displayHeight         240.00
  duration              0.00
  framerate             0.00
  fps                   0.00
  videodatarate         261.00
  videocodecid          7.00
  audiodatarate         62.00
  audiocodecid          10.00
  profile
  level
[live_flv @ 0x1edb0820] Could not find codec parameters for stream 0 (Video: none, none, 267 kb/s): unknown codec
Consider increasing the value for the 'analyzeduration' and 'probesize' options
Input #0, live_flv, from 'rtmp://<server>/input/<stream id>':
  Metadata:
    Server          : NGINX RTMP (github.com/arut/nginx-rtmp-module)
    displayWidth    : 320
    displayHeight   : 240
    fps             : 0
    profile         :
    level           :
  Duration: 00:00:00.00, start: 16.768000, bitrate: N/A
    Stream #0:0: Video: none, none, 267 kb/s, 1k tbr, 1k tbn, 1k tbc
    Stream #0:1: Audio: aac (LC), 16000 Hz, mono, fltp, 63 kb/s
Unsupported codec with id 0 for input stream 0

I am not, by any means, an expert in H264 or video encoding. I know that the encoded frames that come out from MediaCodec contain SPS NAL, PPS NAL, and frame NAL units. I've also written the MediaCodec output into a file and was able to play it back (I did have to specify the format and framerate as otherwise it would play too fast).

My assumption is that things should work (see how little I know :)). Knowing that SPS and PPS are written out, decoder should know enough. Yet, ffprobe fails to recognize codec, fps, and other video information. Do I need to pass packet flag information to FFmpegFrameRecorder.java:record() function? Or should I use direct buffer? Any suggestion will be appreciated! I should figure things out with a hint.

PS: I know that some codecs use Planar and other SemiPlanar color formats. That distinction will come later if I get past this. Also, I didn't go the Surface to MediaCodec way because I need to support API 17 and it requires more changes than this route, which I think helps me understand the more basic flow. Agan, I appreciate any suggestions. Please let me know if something needs to be clarified.

Update #1

So having done more testing, I see that my encoder outputs the following frames:

000000016742800DDA0507E806D0A1350000000168CE06E2
0000000165B840A6F1E7F0EA24000AE73BEB5F51CC7000233A84240...
0000000141E2031364E387FD4F9BB3D67F51CC7000279B9F9CFE811...
0000000141E40304423FFFFF0B7867F89FAFFFFFFFFFFCBE8EF25E6...
0000000141E602899A3512EF8AEAD1379F0650CC3F905131504F839...
...

The very first frame contains SPS and PPS. From what I was able to see, these are transmitted only once. The rest are NAL types 1 and 5. So, my assumption is that, for ffprobe to see stream info not only when the stream starts, I should capture SPS and PPS frames and re-transmit them myself periodically, after a certain number of frames, or perhaps before every I-frame. What do you think?

Update #2

Unable to validate that I'm writing frames successfully. After having tried to read back the written packet, I cannot validate written bytes. As strange, on successful write of IPL image and streaming, I also cannot print out bytes of encoded packet after avcodec_encode_video2. Hit the official dead end.

Answer 1

As far as I know, you don't need to mux video and audio streams. When you send ANNOUNCE message, you also specify to which ports you'll be sending audio and video streams. You need to pack them separately and send with RTP. See wikipedia for more info. https://en.wikipedia.org/wiki/Real_Time_Streaming_Protocol