Recherche avancée

Sur d’autres sites (4141)

• How to grab ffmpeg's output as binary and write it to a file on the fly such that video players can play it in real time ?

  29 décembre 2022, par Mister Mystère

  I want to stream a RTSP-streaming device to a video player such as VLC but the catch is that, in between, the binary data needs to go through a custom high-speed serial link. I control what goes in this link from a C++ program.

  


  I was happily surprised to see that the following line allowed me to watch the RTSP stream by just opening "out.bin" from VLC which was a good lead for fast and efficient binary transmission of the stream :

  


  ffmpeg -i "rtsp://admin:password@X.X.X.X:554/h264Preview_01_main" -c:v copy -c:a copy -f mpegts out.bin


  


  I already wondered how ffmpeg manages to allow VLC to read that file, while itself writing to it at the same time. Turns out I was right to wonder, see below.

  


  I told myself I could make this command pipe its output to the standard output, and then in turn pipe the standard output to a file that I can read, (later, slice it, transmit the chunks and reconstruct it) and then write to an output file. However, this does not work :

  


  #include 
#include 
#include 

#define BUFSIZE 188 //MPEG-TS packet size

int main()
{
    char *cmd = (char*)"ffmpeg -i \"rtsp://admin:password@X.X.X.X:554/h264Preview_01_main\" -c:v copy -c:a copy -f mpegts pipe:1 -loglevel quiet";
    char buf[BUFSIZE];
    FILE *ptr, *file;

    file = fopen("./out.bin", "w");

    if (!file)
    {
        printf("Failed to open output file for writing, aborting");
       abort();
    }

    if ((ptr = popen(cmd, "r")) != NULL) {
       printf("Writing RTSP stream to file...");

       while (!kbhit())
       {
            if(fread(&buf, sizeof(char), BUFSIZE, ptr) != 0)
            {
               fwrite(buf, sizeof(char), BUFSIZE, file);
            }
            else
            {
                printf("No data\n");
            }
       }
       pclose(ptr);
    }
    else
    {
        printf("Failed to open pipe from ffmpeg command, aborting");
    }

    printf("End of program");

    fclose(file);
    return 0;
}


  


  Since VLC says "your input can't be opened" - whereas this works just fine :

  


  ffmpeg -i "rtsp://admin:password@X.X.X.X:554/h264Preview_01_main" -c:v copy -c:a copy -f mpegts pipe:1 -loglevel quiet > out.bin


  


  This is what ends up in the file after I close the program, versus the result of the command immediately above :


  


  The file is always 2kB regardless of how long I run the program : "No data" is shown repeatedly in the console output.

  


  Why doesn't it work ? If it is not just a bug, how can I grab the stream as binary at some point, and write it at the end to a file that VLC can read ?

  


  Update

  


  New code after applying Craig Estey's fix to my stupid mistake. The end result is that the MPEG-TS frames don't seem to shift anymore but the file writing stops partway into one of the first few frames (the console only shows a few ">" symbols and then stays silent, c.f. code).

  


  #include 
#include 
#include 

#define BUFSIZE 188                     // MPEG-TS packet size

int
main()
{
    char *cmd = (char *) "ffmpeg -i \"rtsp://127.0.0.1:8554/test.sdp\" -c:v copy -c:a copy -f mpegts pipe:1 -loglevel quiet";
    char buf[BUFSIZE];
    FILE *ptr,
    *file;

    file = fopen("./out.ts", "w");

    if (!file) {
        printf("Failed to open output file for writing, aborting");
        abort();
    }

    if ((ptr = popen(cmd, "r")) != NULL) {
        printf("Writing RTSP stream to file...");

        while(!kbhit()) {
            ssize_t rlen = fread(&buf, sizeof(char), BUFSIZE, ptr);
            if(rlen != 0)
            {
                printf(">");
                fwrite(buf, sizeof(char), rlen, file);
                fflush(file);
            }
        }
        pclose(ptr);
    }
    else {
        printf("Failed to open pipe from ffmpeg command, aborting");
    }

    printf("End of program");

    fclose(file);
    return 0;
}


  


  This can be tested on any computer with VLC and a webcam : open VLC, open capture device, capture mode directshow, (switch "play" for "stream"), next, display locally, select RTSP, Add, path=/test.sdp, next, transcoding=H264+MP3 (TS), replace rtsp ://:8554/ with rtsp ://127.0.0.1:8554/ in the generated command line, stream.

  


  To test that streaming is ok, you can just open a command terminal and enter "ffmpeg -i "rtsp ://127.0.0.1:8554/test.sdp" -c:v copy -c:a copy -f mpegts pipe:1 -loglevel quiet", the terminal should fill up with binary data.

  


  To test the program, just compile, run, and open out.ts after the program has run.

  


• How to feed frames one by one and grab decoded frames using LibAV ?

  15 septembre 2024, par Alvan Rahimli

  I am integrating a third party system where I send a TCP packet to ask e.g. 5 frames from a live CCTV camera stream, and it sends those 5 frames one by one. Each package is h264 encoded frame, wrapped with some relevant data.

  


  I need to write a code using Libav where I can :

  


  

  1. Feed the frames one by one using AVIOContext (or smth similar).

  


  2. Grab decoded frame.

  


  3. Draw that to a window (Not relevant to the question, writing for context).

  


  


  I've been doing the same with GStreamer by creating a pipeline like this :

  


  AppSrc -> H264Parser -> H264Decoder -> FrameGrabber


  


  The code below is what I was able to write so far :

  


  using System.Runtime.InteropServices;
using FFmpeg.AutoGen;

namespace AvIoCtxExperiment;

public static unsafe class Program
{
    public static void Main()
    {
        ffmpeg.avdevice_register_all();
        Console.WriteLine($"FFmpeg v: {ffmpeg.av_version_info()}");

        // Generous buffer size for I frames (~43KB)
        const int bufferSize = 50 * 1024;

        var buff = (byte*)ffmpeg.av_malloc(bufferSize);
        if (buff == null)
            throw new Exception("Buffer is null");

        // A mock frame provider. Frames are stored in separate files and this reads & returns them one-by-one.
        var frameProvider = new FrameProvider(@"D:\Frames-1", 700);
        var gch = GCHandle.Alloc(frameProvider);

        var avioCtx = ffmpeg.avio_alloc_context(
            buffer: buff,
            buffer_size: bufferSize,
            write_flag: 0,
            opaque: (void*)GCHandle.ToIntPtr(gch),
            read_packet: new avio_alloc_context_read_packet(ReadFunc),
            write_packet: null,
            seek: null);

        var formatContext = ffmpeg.avformat_alloc_context();
        formatContext->pb = avioCtx;
        formatContext->flags |= ffmpeg.AVFMT_FLAG_CUSTOM_IO;

        var openResult = ffmpeg.avformat_open_input(&formatContext, null, null, null);
        if (openResult < 0)
            throw new Exception("Open Input Failed");

        if (ffmpeg.avformat_find_stream_info(formatContext, null) < 0)
            throw new Exception("Find StreamInfo Failed");

        AVPacket packet;
        while (ffmpeg.av_read_frame(formatContext, &packet) >= 0)
        {
            Console.WriteLine($"GRAB: {packet.buf->size}");
            ffmpeg.av_packet_unref(&packet);
        }
    }

    private static int ReadFunc(void* opaque, byte* buf, int bufSize)
    {
        var frameProvider = (FrameProvider?)GCHandle.FromIntPtr((IntPtr)opaque).Target;
        if (frameProvider == null)
        {
            return 0;
        }

        byte[] managedBuffer = new byte[bufSize];

        var fBuff = frameProvider.NextFrame();
        if (fBuff == null)
        {
            return ffmpeg.AVERROR_EOF;
        }

        int bytesRead = fBuff.Length;
        fBuff.CopyTo(managedBuffer, 0);

        if (bytesRead > 0)
        {
            Marshal.Copy(managedBuffer, 0, (IntPtr)buf, bytesRead);
            Console.WriteLine($"READ size: {fBuff.Length}");
            return bytesRead;
        }

        return ffmpeg.AVERROR_EOF;
    }
}


  


  Second thing that confuses me is with AppSrc, I can feed frame of any size, but with LibAV, it asks for buffer size. With AppSrc, I am responsible of feeding frames into the pipeline. GStreamer notifies me when it is enough with signals. But with libav, it calls read_packet delegate itself.

  


  Any help is much appreciated. Thanks.

  


  P.S. I'm writing a C# app, but sample code in C is fine, I can adapt the code myself.

  


• How to feed frames one by one and grab decoded frames using LibAV ?

  22 septembre 2024, par Alvan Rahimli

  I am integrating a third party system where I send a TCP packet to ask e.g. 5 frames from a live CCTV camera stream, and it sends those 5 frames one by one. Each package is h264 encoded frame, wrapped with some relevant data.

  


  I need to write a code using Libav where I can :

  


  

  1. Feed the frames one by one using AVIOContext (or smth similar).

  


  2. Grab decoded frame.

  


  3. Draw that to a window (Not relevant to the question, writing for context).

  


  


  I've been doing the same with GStreamer by creating a pipeline like this :

  


  AppSrc -> H264Parser -> H264Decoder -> FrameGrabber


  


  The code below is what I was able to write so far :

  


  using System.Runtime.InteropServices;
using FFmpeg.AutoGen;

namespace AvIoCtxExperiment;

public static unsafe class Program
{
    public static void Main()
    {
        ffmpeg.avdevice_register_all();
        Console.WriteLine($"FFmpeg v: {ffmpeg.av_version_info()}");

        // Generous buffer size for I frames (~43KB)
        const int bufferSize = 50 * 1024;

        var buff = (byte*)ffmpeg.av_malloc(bufferSize);
        if (buff == null)
            throw new Exception("Buffer is null");

        // A mock frame provider. Frames are stored in separate files and this reads & returns them one-by-one.
        var frameProvider = new FrameProvider(@"D:\Frames-1", 700);
        var gch = GCHandle.Alloc(frameProvider);

        var avioCtx = ffmpeg.avio_alloc_context(
            buffer: buff,
            buffer_size: bufferSize,
            write_flag: 0,
            opaque: (void*)GCHandle.ToIntPtr(gch),
            read_packet: new avio_alloc_context_read_packet(ReadFunc),
            write_packet: null,
            seek: null);

        var formatContext = ffmpeg.avformat_alloc_context();
        formatContext->pb = avioCtx;
        formatContext->flags |= ffmpeg.AVFMT_FLAG_CUSTOM_IO;

        var openResult = ffmpeg.avformat_open_input(&formatContext, null, null, null);
        if (openResult < 0)
            throw new Exception("Open Input Failed");

        if (ffmpeg.avformat_find_stream_info(formatContext, null) < 0)
            throw new Exception("Find StreamInfo Failed");

        AVPacket packet;
        while (ffmpeg.av_read_frame(formatContext, &packet) >= 0)
        {
            Console.WriteLine($"GRAB: {packet.buf->size}");
            ffmpeg.av_packet_unref(&packet);
        }
    }

    private static int ReadFunc(void* opaque, byte* buf, int bufSize)
    {
        var frameProvider = (FrameProvider?)GCHandle.FromIntPtr((IntPtr)opaque).Target;
        if (frameProvider == null)
        {
            return 0;
        }

        byte[] managedBuffer = new byte[bufSize];

        var fBuff = frameProvider.NextFrame();
        if (fBuff == null)
        {
            return ffmpeg.AVERROR_EOF;
        }

        int bytesRead = fBuff.Length;
        fBuff.CopyTo(managedBuffer, 0);

        if (bytesRead > 0)
        {
            Marshal.Copy(managedBuffer, 0, (IntPtr)buf, bytesRead);
            Console.WriteLine($"READ size: {fBuff.Length}");
            return bytesRead;
        }

        return ffmpeg.AVERROR_EOF;
    }
}


  


  Second thing that confuses me is with AppSrc, I can feed frame of any size, but with LibAV, it asks for buffer size. With AppSrc, I am responsible of feeding frames into the pipeline. GStreamer notifies me when it is enough with signals. But with libav, it calls read_packet delegate itself.

  


  Any help is much appreciated. Thanks.

  


  P.S. I'm writing a C# app, but sample code in C is fine, I can adapt the code myself.