Recherche avancée

Sur d’autres sites (5871)

• aacpsy : correct calculation of minath in psy_3gpp_init

  21 avril 2015, par Andreas Cadhalpun

  aacpsy : correct calculation of minath in psy_3gpp_init
  The minimum of the ath(x, ATH_ADD) function depends on ATH_ADD.
  
  This patch uses the first order approximation to determine it.
  For ATH_ADD = 4 this results in the value at 3407.06812 (-5.24241638)
  
  not the one at 3410 (-5.24237967).
  Signed-off-by : Andreas Cadhalpun <Andreas.Cadhalpun@googlemail.com>
  
  Approved-by : Claudio Freire <klaussfreire@gmail.com>
  
  Signed-off-by : Michael Niedermayer <michaelni@gmx.at>
  

  • [DH] libavcodec/aacpsy.c

• FFmpeg C Api - Reduce fps but maintain video duration

  25 mars 2015, par Justin Bradley

  Using the FFmpeg C API I’m trying to convert an input video into a video that looks like an animated gif - meaning no audio stream and a video stream of 4/fps.

  I have the decode/encode part working. I can drop the audio stream from the output file, but I’m having trouble reducing the fps. I can change the output video stream’s time_base to 4/fps, but it increases the video’s duration - basically playing it in slow mo.

  I think I need to drop the extra frames before I write them to the output container.

  Below is the loop where I read the input frames, and then write them to output container.

  Is this where I’d drop the extra frames ? How do I determine which frames to drop (I,P,B frames) ?

  while(av_read_frame(input_container, &decoded_packet)>=0) {
  
  
  
      if (decoded_packet.stream_index == video_stream_index) {
  
          len = avcodec_decode_video2(input_stream->codec, decoded_frame, &got_frame, &decoded_packet);
  
          if(len < 0) {
  
              exit(1);
  
          }
  
  
  
          if(got_frame) {
  
              av_init_packet(&encoded_packet);
  
              encoded_packet.data =  NULL;
  
              encoded_packet.size =  0;
  
  
  
              if(avcodec_encode_video2(output_stream->codec, &encoded_packet, decoded_frame, &got_frame) < 0) {
  
                  exit(1);
  
              }
  
              if(got_frame) {
  
                  if (output_stream->codec->coded_frame->key_frame) {
  
                      encoded_packet.flags |= AV_PKT_FLAG_KEY;
  
                  }
  
  
  
                  encoded_packet.stream_index = output_stream->index;
  
                  encoded_packet.pts = av_rescale_q(current_frame_num, output_stream->codec->time_base, output_stream->time_base);
  
                  encoded_packet.dts = av_rescale_q(current_frame_num, output_stream->codec->time_base, output_stream->time_base);
  
  
  
                  if(av_interleaved_write_frame(output_container, &encoded_packet) < 0) {
  
                      exit(1);
  
                  }
  
                  else {
  
                      current_frame_num +=1;
  
                  }
  
              }
  
              frame_count+=1;
  
              av_free_packet(&encoded_packet);
  
          }
  
      }
  
  }

• How to stream synchronized video and audio in real-time from an Android smartphone using HLS while preserving orientation metadata ?

  6 mars, par Jérôme LAROSE

  Hello,  
I am working on an Android application where I need to stream video
from one or two cameras on my smartphone, along with audio from the
microphone, in real-time via a link or web page accessible to users.
The stream should be live, allow rewinding (DVR functionality), and be
recorded simultaneously. A latency of 1 to 2 minutes is acceptable,
and the streaming is one-way.  

I have chosen HLS (HTTP Live Streaming) for its browser compatibility
and DVR support. However, I am encountering issues with audio-video
synchronization, managing camera orientation metadata, and format
conversions.


  


  Here are my attempts :

  


  


  1. MP4 segmentation with MediaRecorder

     


     

     • I used MediaRecorder with setNextOutputFile to generate short MP4 segments, then ffmpeg-kit to convert them to fMP4 for HLS.

     


     • Expected : Well-aligned segments for smooth HLS playback.

     


     • Result : Timestamp issues causing jumps or interruptions in playback.

     


     


  


  2. MPEG2-TS via local socket

     


     

     • I configured MediaRecorder to produce an MPEG2-TS stream sent via a local socket to ffmpeg-kit.

     


     • Expected : Stable streaming with preserved metadata.

     


     • Result : Streaming works, but orientation metadata is lost, leading to incorrectly oriented video (e.g., rotated 90°).

     


     


  


  3. Orientation correction with ffmpeg

     


     

     • I tested -vf transpose=1 in ffmpeg to correct the orientation.

     


     • Expected : Correctly oriented video without excessive latency.

     


     • Result : Re-encoding takes too long for real-time streaming, causing unacceptable latency.

     


     


  


  4. MPEG2-TS to fMP4 conversion

     


     

     • I converted the MPEG2-TS stream to fMP4 with ffmpeg to preserve orientation.

     


     • Expected : Perfect audio-video synchronization.

     


     • Result : Slight desynchronization between audio and video, affecting the user experience.

     


     


  


  


  I am looking for a solution to :

  


  

  • Stream an HLS feed from Android with correctly timestamped segments.

  


  • Preserve orientation metadata without heavy re-encoding.

  


  • Ensure perfect audio-video synchronization.

  


  


  UPDATE

  


  package com.example.angegardien

import android.Manifest
import android.content.Context
import android.content.pm.PackageManager
import android.graphics.SurfaceTexture
import android.hardware.camera2.*
import android.media.*
import android.os.*
import android.util.Log
import android.view.Surface
import android.view.TextureView
import android.view.WindowManager
import androidx.activity.ComponentActivity
import androidx.core.app.ActivityCompat
import com.arthenica.ffmpegkit.FFmpegKit
import fi.iki.elonen.NanoHTTPD
import kotlinx.coroutines.*
import java.io.File
import java.io.IOException
import java.net.ServerSocket
import android.view.OrientationEventListener

/**
 * MainActivity class:
 * - Manages camera operations using the Camera2 API.
 * - Records video using MediaRecorder.
 * - Pipes data to FFmpeg to generate HLS segments.
 * - Hosts a local HLS server using NanoHTTPD to serve the generated HLS content.
 */
class MainActivity : ComponentActivity() {

    // TextureView used for displaying the camera preview.
    private lateinit var textureView: TextureView
    // Camera device instance.
    private lateinit var cameraDevice: CameraDevice
    // Camera capture session for managing capture requests.
    private lateinit var cameraCaptureSession: CameraCaptureSession
    // CameraManager to access camera devices.
    private lateinit var cameraManager: CameraManager
    // Directory where HLS output files will be stored.
    private lateinit var hlsDir: File
    // Instance of the HLS server.
    private lateinit var hlsServer: HlsServer

    // Camera id ("1" corresponds to the rear camera).
    private val cameraId = "1"
    // Flag indicating whether recording is currently active.
    private var isRecording = false

    // MediaRecorder used for capturing audio and video.
    private lateinit var activeRecorder: MediaRecorder
    // Surface for the camera preview.
    private lateinit var previewSurface: Surface
    // Surface provided by MediaRecorder for recording.
    private lateinit var recorderSurface: Surface

    // Port for the FFmpeg local socket connection.
    private val ffmpegPort = 8080

    // Coroutine scope to manage asynchronous tasks.
    private val scope = CoroutineScope(Dispatchers.IO + SupervisorJob())

    // Variables to track current device rotation and listen for orientation changes.
    private var currentRotation = 0
    private lateinit var orientationListener: OrientationEventListener

    override fun onCreate(savedInstanceState: Bundle?) {
        super.onCreate(savedInstanceState)

        // Initialize the TextureView and set it as the content view.
        textureView = TextureView(this)
        setContentView(textureView)

        // Get the CameraManager system service.
        cameraManager = getSystemService(CAMERA_SERVICE) as CameraManager
        // Setup the directory for HLS output.
        setupHLSDirectory()

        // Start the local HLS server on port 8081.
        hlsServer = HlsServer(8081, hlsDir, this)
        try {
            hlsServer.start()
            Log.d("HLS_SERVER", "HLS Server started on port 8081")
        } catch (e: IOException) {
            Log.e("HLS_SERVER", "Error starting HLS Server", e)
        }

        // Initialize the current rotation.
        currentRotation = getDeviceRotation()

        // Add a listener to detect orientation changes.
        orientationListener = object : OrientationEventListener(this) {
            override fun onOrientationChanged(orientation: Int) {
                if (orientation == ORIENTATION_UNKNOWN) return // Skip unknown orientations.
                // Determine the new rotation angle.
                val newRotation = when {
                    orientation >= 315 || orientation < 45 -> 0
                    orientation >= 45 && orientation < 135 -> 90
                    orientation >= 135 && orientation < 225 -> 180
                    orientation >= 225 && orientation < 315 -> 270
                    else -> 0
                }
                // If the rotation has changed and recording is active, update the rotation.
                if (newRotation != currentRotation && isRecording) {
                    Log.d("ROTATION", "Orientation change detected: $newRotation")
                    currentRotation = newRotation
                }
            }
        }
        orientationListener.enable()

        // Set up the TextureView listener to know when the surface is available.
        textureView.surfaceTextureListener = object : TextureView.SurfaceTextureListener {
            override fun onSurfaceTextureAvailable(surface: SurfaceTexture, width: Int, height: Int) {
                // Open the camera when the texture becomes available.
                openCamera()
            }
            override fun onSurfaceTextureSizeChanged(surface: SurfaceTexture, width: Int, height: Int) {}
            override fun onSurfaceTextureDestroyed(surface: SurfaceTexture) = false
            override fun onSurfaceTextureUpdated(surface: SurfaceTexture) {}
        }
    }

    /**
     * Sets up the HLS directory in the public Downloads folder.
     * If the directory exists, it deletes it recursively and creates a new one.
     */
    private fun setupHLSDirectory() {
        val downloadsDir = Environment.getExternalStoragePublicDirectory(Environment.DIRECTORY_DOWNLOADS)
        hlsDir = File(downloadsDir, "HLS_Output")

        if (hlsDir.exists()) {
            hlsDir.deleteRecursively()
        }
        hlsDir.mkdirs()

        Log.d("HLS", "📂 HLS folder created: ${hlsDir.absolutePath}")
    }

    /**
     * Opens the camera after checking for necessary permissions.
     */
    private fun openCamera() {
        if (ActivityCompat.checkSelfPermission(this, Manifest.permission.CAMERA) != PackageManager.PERMISSION_GRANTED ||
            ActivityCompat.checkSelfPermission(this, Manifest.permission.RECORD_AUDIO) != PackageManager.PERMISSION_GRANTED) {
            // Request permissions if they are not already granted.
            ActivityCompat.requestPermissions(this, arrayOf(Manifest.permission.CAMERA, Manifest.permission.RECORD_AUDIO), 101)
            return
        }

        try {
            // Open the specified camera using its cameraId.
            cameraManager.openCamera(cameraId, object : CameraDevice.StateCallback() {
                override fun onOpened(camera: CameraDevice) {
                    cameraDevice = camera
                    // Start the recording session once the camera is opened.
                    startNextRecording()
                }
                override fun onDisconnected(camera: CameraDevice) { camera.close() }
                override fun onError(camera: CameraDevice, error: Int) { camera.close() }
            }, null)
        } catch (e: CameraAccessException) {
            e.printStackTrace()
        }
    }

    /**
     * Starts a new recording session:
     * - Sets up the preview and recorder surfaces.
     * - Creates a pipe for MediaRecorder output.
     * - Creates a capture session for simultaneous preview and recording.
     */
    private fun startNextRecording() {
        // Get the SurfaceTexture from the TextureView and set its default buffer size.
        val texture = textureView.surfaceTexture!!
        texture.setDefaultBufferSize(1920, 1080)
        // Create the preview surface.
        previewSurface = Surface(texture)

        // Create and configure the MediaRecorder.
        activeRecorder = createMediaRecorder()

        // Create a pipe to route MediaRecorder data.
        val pipe = ParcelFileDescriptor.createPipe()
        val pfdWrite = pipe[1] // Write end used by MediaRecorder.
        val pfdRead = pipe[0]  // Read end used by the local socket server.

        // Set MediaRecorder output to the file descriptor of the write end.
        activeRecorder.setOutputFile(pfdWrite.fileDescriptor)
        setupMediaRecorder(activeRecorder)
        // Obtain the recorder surface from MediaRecorder.
        recorderSurface = activeRecorder.surface

        // Create a capture request using the RECORD template.
        val captureRequestBuilder = cameraDevice.createCaptureRequest(CameraDevice.TEMPLATE_RECORD)
        captureRequestBuilder.addTarget(previewSurface)
        captureRequestBuilder.addTarget(recorderSurface)

        // Create a capture session including both preview and recorder surfaces.
        cameraDevice.createCaptureSession(
            listOf(previewSurface, recorderSurface),
            object : CameraCaptureSession.StateCallback() {
                override fun onConfigured(session: CameraCaptureSession) {
                    cameraCaptureSession = session
                    captureRequestBuilder.set(CaptureRequest.CONTROL_MODE, CameraMetadata.CONTROL_MODE_AUTO)
                    // Start a continuous capture request.
                    cameraCaptureSession.setRepeatingRequest(captureRequestBuilder.build(), null, null)

                    // Launch a coroutine to start FFmpeg and MediaRecorder with synchronization.
                    scope.launch {
                        startFFmpeg()
                        delay(500) // Wait for FFmpeg to be ready.
                        activeRecorder.start()
                        isRecording = true
                        Log.d("HLS", "🎥 Recording started...")
                    }

                    // Launch a coroutine to run the local socket server to forward data.
                    scope.launch {
                        startLocalSocketServer(pfdRead)
                    }
                }
                override fun onConfigureFailed(session: CameraCaptureSession) {
                    Log.e("Camera2", "❌ Configuration failed")
                }
            },
            null
        )
    }

    /**
     * Coroutine to start a local socket server.
     * It reads from the MediaRecorder pipe and sends the data to FFmpeg.
     */
    private suspend fun startLocalSocketServer(pfdRead: ParcelFileDescriptor) {
        withContext(Dispatchers.IO) {
            val serverSocket = ServerSocket(ffmpegPort)
            Log.d("HLS", "Local socket server started on port $ffmpegPort")

            // Accept connection from FFmpeg.
            val socket = serverSocket.accept()
            Log.d("HLS", "Connection accepted from FFmpeg")

            // Read data from the pipe and forward it through the socket.
            val inputStream = ParcelFileDescriptor.AutoCloseInputStream(pfdRead)
            val outputStream = socket.getOutputStream()
            val buffer = ByteArray(8192)
            var bytesRead: Int
            while (inputStream.read(buffer).also { bytesRead = it } != -1) {
                outputStream.write(buffer, 0, bytesRead)
            }
            outputStream.close()
            inputStream.close()
            socket.close()
            serverSocket.close()
        }
    }

    /**
     * Coroutine to start FFmpeg using a local TCP input.
     * Applies a video rotation filter based on device orientation and generates HLS segments.
     */
    private suspend fun startFFmpeg() {
        withContext(Dispatchers.IO) {
            // Retrieve the appropriate transpose filter based on current rotation.
            val transposeFilter = getTransposeFilter(currentRotation)

            // FFmpeg command to read from the TCP socket and generate an HLS stream.
            // Two alternative commands are commented below.
            // val ffmpegCommand = "-fflags +genpts -i tcp://localhost:$ffmpegPort -c copy -bsf:a aac_adtstoasc -movflags +faststart -f dash -seg_duration 10 -hls_playlist 1 ${hlsDir.absolutePath}/manifest.mpd"
            // val ffmpegCommand = "-fflags +genpts -i tcp://localhost:$ffmpegPort -c copy -bsf:a aac_adtstoasc -movflags +faststart -f hls -hls_time 5 -hls_segment_type fmp4 -hls_flags split_by_time -hls_list_size 0 -hls_playlist_type event -hls_fmp4_init_filename init.mp4 -hls_segment_filename ${hlsDir.absolutePath}/segment_%03d.m4s ${hlsDir.absolutePath}/playlist.m3u8"
            val ffmpegCommand = "-fflags +genpts -i tcp://localhost:$ffmpegPort -vf $transposeFilter -c:v libx264 -preset ultrafast -crf 23 -c:a copy -movflags +faststart -f hls -hls_time 0.1 -hls_segment_type mpegts -hls_flags split_by_time -hls_list_size 0 -hls_playlist_type event -hls_segment_filename ${hlsDir.absolutePath}/segment_%03d.ts ${hlsDir.absolutePath}/playlist.m3u8"

            FFmpegKit.executeAsync(ffmpegCommand) { session ->
                if (session.returnCode.isValueSuccess) {
                    Log.d("HLS", "✅ HLS generated successfully")
                } else {
                    Log.e("FFmpeg", "❌ Error generating HLS: ${session.allLogsAsString}")
                }
            }
        }
    }

    /**
     * Gets the current device rotation using the WindowManager.
     */
    private fun getDeviceRotation(): Int {
        val windowManager = getSystemService(Context.WINDOW_SERVICE) as WindowManager
        return when (windowManager.defaultDisplay.rotation) {
            Surface.ROTATION_0 -> 0
            Surface.ROTATION_90 -> 90
            Surface.ROTATION_180 -> 180
            Surface.ROTATION_270 -> 270
            else -> 0
        }
    }

    /**
     * Returns the FFmpeg transpose filter based on the rotation angle.
     * Used to rotate the video stream accordingly.
     */
    private fun getTransposeFilter(rotation: Int): String {
        return when (rotation) {
            90 -> "transpose=1" // 90° clockwise
            180 -> "transpose=2,transpose=2" // 180° rotation
            270 -> "transpose=2" // 90° counter-clockwise
            else -> "transpose=0" // No rotation
        }
    }

    /**
     * Creates and configures a MediaRecorder instance.
     * Sets up audio and video sources, formats, encoders, and bitrates.
     */
    private fun createMediaRecorder(): MediaRecorder {
        return MediaRecorder().apply {
            setAudioSource(MediaRecorder.AudioSource.MIC)
            setVideoSource(MediaRecorder.VideoSource.SURFACE)
            setOutputFormat(MediaRecorder.OutputFormat.MPEG_2_TS)
            setVideoEncodingBitRate(5000000)
            setVideoFrameRate(24)
            setVideoSize(1080, 720)
            setVideoEncoder(MediaRecorder.VideoEncoder.H264)
            setAudioEncoder(MediaRecorder.AudioEncoder.AAC)
            setAudioSamplingRate(16000)
            setAudioEncodingBitRate(96000) // 96 kbps
        }
    }

    /**
     * Prepares the MediaRecorder and logs the outcome.
     */
    private fun setupMediaRecorder(recorder: MediaRecorder) {
        try {
            recorder.prepare()
            Log.d("HLS", "✅ MediaRecorder prepared")
        } catch (e: IOException) {
            Log.e("HLS", "❌ Error preparing MediaRecorder", e)
        }
    }

    /**
     * Custom HLS server class extending NanoHTTPD.
     * Serves HLS segments and playlists from the designated HLS directory.
     */
    private inner class HlsServer(port: Int, private val hlsDir: File, private val context: Context) : NanoHTTPD(port) {
        override fun serve(session: IHTTPSession): Response {
            val uri = session.uri.trimStart('/')

            // Intercept the request for `init.mp4` and serve it from assets.
            /*
            if (uri == "init.mp4") {
                Log.d("HLS Server", "📡 Intercepting init.mp4, sending file from assets...")
                return try {
                    val assetManager = context.assets
                    val inputStream = assetManager.open("init.mp4")
                    newFixedLengthResponse(Response.Status.OK, "video/mp4", inputStream, inputStream.available().toLong())
                } catch (e: Exception) {
                    Log.e("HLS Server", "❌ Error reading init.mp4 from assets: ${e.message}")
                    newFixedLengthResponse(Response.Status.INTERNAL_ERROR, MIME_PLAINTEXT, "Server error")
                }
            }
            */

            // Serve all other HLS files normally from the hlsDir.
            val file = File(hlsDir, uri)
            return if (file.exists()) {
                newFixedLengthResponse(Response.Status.OK, getMimeTypeForFile(uri), file.inputStream(), file.length())
            } else {
                newFixedLengthResponse(Response.Status.NOT_FOUND, MIME_PLAINTEXT, "File not found")
            }
        }
    }

    /**
     * Clean up resources when the activity is destroyed.
     * Stops recording, releases the camera, cancels coroutines, and stops the HLS server.
     */
    override fun onDestroy() {
        super.onDestroy()
        if (isRecording) {
            activeRecorder.stop()
            activeRecorder.release()
        }
        cameraDevice.close()
        scope.cancel()
        hlsServer.stop()
        orientationListener.disable()
        Log.d("HLS", "🛑 Activity destroyed")
    }
}


  


  I have three examples of ffmpeg commands.

  


  

  • One command segments into DASH, but the camera does not have the correct rotation.

  


  • One command segments into HLS without re-encoding with 5-second segments ; it’s fast but does not have the correct rotation.

  


  • One command segments into HLS with re-encoding, which applies a rotation. It’s too slow for 5-second segments, so a 1-second segment was chosen.

  


  


  Note :

  


  

  • In the second command ("One command segments into HLS without re-encoding with 5-second segments ; it’s fast but does not have the correct rotation."), it returns fMP4. To achieve the correct rotation, I provide a preconfigured init.mp4 file during the HTTP request to retrieve it (see comment).

  


  • In the third command ("One command segments into HLS with re-encoding, which applies a rotation. It’s too slow for 5-second segments, so a 1-second segment was chosen."), it returns TS.