java - 无法理解这种方法(它如何尝试匹配帧率？)

Question

我在浏览 the tutorial on how to decode a video 时看到了这段代码:

private static long millisecondsUntilTimeToDisplay(IVideoPicture picture)
{
/**
 * We could just display the images as quickly as we decode them, but it turns
 * out we can decode a lot faster than you think.
 * 
 * So instead, the following code does a poor-man's version of trying to
 * match up the frame-rate requested for each IVideoPicture with the system
 * clock time on your computer.
 * 
 * Remember that all Xuggler IAudioSamples and IVideoPicture objects always
 * give timestamps in Microseconds, relative to the first decoded item.  If
 * instead you used the packet timestamps, they can be in different units depending
 * on your IContainer, and IStream and things can get hairy quickly.
 */
long millisecondsToSleep = 0;
if (mFirstVideoTimestampInStream == Global.NO_PTS)
{
  // This is our first time through
  mFirstVideoTimestampInStream = picture.getTimeStamp();
  // get the starting clock time so we can hold up frames
  // until the right time.
  mSystemVideoClockStartTime = System.currentTimeMillis();
  millisecondsToSleep = 0;
} else {
  long systemClockCurrentTime = System.currentTimeMillis();
  long millisecondsClockTimeSinceStartofVideo = systemClockCurrentTime - mSystemVideoClockStartTime;
  // compute how long for this frame since the first frame in the stream.
  // remember that IVideoPicture and IAudioSamples timestamps are always in MICROSECONDS,
  // so we divide by 1000 to get milliseconds.
  long millisecondsStreamTimeSinceStartOfVideo = (picture.getTimeStamp() - mFirstVideoTimestampInStream)/1000;
  final long millisecondsTolerance = 50; // and we give ourselfs 50 ms of tolerance
  millisecondsToSleep = (millisecondsStreamTimeSinceStartOfVideo -
      (millisecondsClockTimeSinceStartofVideo+millisecondsTolerance));
}
return millisecondsToSleep;

我已经抓了很多但不明白这个方法是做什么的？我们要返回什么？ 为什么我们要在方法返回后让线程 hibernate (这个方法的目的是什么？)

这是链接中的完整代码:

import javax.sound.sampled.AudioFormat;
import javax.sound.sampled.AudioSystem;
import javax.sound.sampled.DataLine;
import javax.sound.sampled.LineUnavailableException;
import javax.sound.sampled.SourceDataLine;
import com.xuggle.xuggler.demos.*;
import com.xuggle.xuggler.Global;
import com.xuggle.xuggler.IAudioSamples;
import com.xuggle.xuggler.IContainer;
import com.xuggle.xuggler.IPacket;
import com.xuggle.xuggler.IPixelFormat;
import com.xuggle.xuggler.IStream;
import com.xuggle.xuggler.IStreamCoder;
import com.xuggle.xuggler.ICodec;
import com.xuggle.xuggler.IVideoPicture;
import com.xuggle.xuggler.IVideoResampler;
import com.xuggle.xuggler.Utils;

public class DecodeAndPlayAudioAndVideo
{

/**
 * The audio line we'll output sound to; it'll be the default audio device on your      system if available
*/
private static SourceDataLine mLine;

/**
 * The window we'll draw the video on.
 * 
 */
 private static VideoImage mScreen = null;

 private static long mSystemVideoClockStartTime;

 private static long mFirstVideoTimestampInStream;

 /**
  * Takes a media container (file) as the first argument, opens it,
  * plays audio as quickly as it can, and opens up a Swing window and displays
  * video frames with <i>roughly</i> the right timing.
  *  
  * @param args Must contain one string which represents a filename
  */
  @SuppressWarnings("deprecation")
  public static void main(String[] args)
  {
if (args.length <= 0)
  throw new IllegalArgumentException("must pass in a filename as the first argument");

String filename = args[0];

// Let's make sure that we can actually convert video pixel formats.
if (!IVideoResampler.isSupported(IVideoResampler.Feature.FEATURE_COLORSPACECONVERSION))
  throw new RuntimeException("you must install the GPL version of Xuggler (with IVideoResampler support) for this demo to work");

// Create a Xuggler container object
IContainer container = IContainer.make();

// Open up the container
if (container.open(filename, IContainer.Type.READ, null) < 0)
  throw new IllegalArgumentException("could not open file: " + filename);

// query how many streams the call to open found
int numStreams = container.getNumStreams();

// and iterate through the streams to find the first audio stream
int videoStreamId = -1;
IStreamCoder videoCoder = null;
int audioStreamId = -1;
IStreamCoder audioCoder = null;
for(int i = 0; i < numStreams; i++)
{
  // Find the stream object
  IStream stream = container.getStream(i);
  // Get the pre-configured decoder that can decode this stream;
  IStreamCoder coder = stream.getStreamCoder();

  if (videoStreamId == -1 && coder.getCodecType() == ICodec.Type.CODEC_TYPE_VIDEO)
  {
    videoStreamId = i;
    videoCoder = coder;
  }
  else if (audioStreamId == -1 && coder.getCodecType() == ICodec.Type.CODEC_TYPE_AUDIO)
  {
    audioStreamId = i;
    audioCoder = coder;
  }
}
if (videoStreamId == -1 && audioStreamId == -1)
  throw new RuntimeException("could not find audio or video stream in container: "+filename);

/*
 * Check if we have a video stream in this file.  If so let's open up our decoder so it can
 * do work.
 */
IVideoResampler resampler = null;
if (videoCoder != null)
{
  if(videoCoder.open() < 0)
    throw new RuntimeException("could not open audio decoder for container: "+filename);

  if (videoCoder.getPixelType() != IPixelFormat.Type.BGR24)
  {
    // if this stream is not in BGR24, we're going to need to
    // convert it.  The VideoResampler does that for us.
    resampler = IVideoResampler.make(videoCoder.getWidth(), videoCoder.getHeight(), IPixelFormat.Type.BGR24,
        videoCoder.getWidth(), videoCoder.getHeight(), videoCoder.getPixelType());
    if (resampler == null)
      throw new RuntimeException("could not create color space resampler for: " + filename);
  }
  /*
   * And once we have that, we draw a window on screen
   */
  openJavaVideo();
}

if (audioCoder != null)
{
  if (audioCoder.open() < 0)
    throw new RuntimeException("could not open audio decoder for container: "+filename);

  /*
   * And once we have that, we ask the Java Sound System to get itself ready.
   */
  try
  {
    openJavaSound(audioCoder);
  }
  catch (LineUnavailableException ex)
  {
    throw new RuntimeException("unable to open sound device on your system when playing back container: "+filename);
  }
}


/*
 * Now, we start walking through the container looking at each packet.
 */
IPacket packet = IPacket.make();
mFirstVideoTimestampInStream = Global.NO_PTS;
mSystemVideoClockStartTime = 0;
while(container.readNextPacket(packet) >= 0)
{
  /*
   * Now we have a packet, let's see if it belongs to our video stream
   */
  if (packet.getStreamIndex() == videoStreamId)
  {
    /*
     * We allocate a new picture to get the data out of Xuggler
     */
    IVideoPicture picture = IVideoPicture.make(videoCoder.getPixelType(),
        videoCoder.getWidth(), videoCoder.getHeight());

    /*
     * Now, we decode the video, checking for any errors.
     * 
     */
    int bytesDecoded = videoCoder.decodeVideo(picture, packet, 0);
    if (bytesDecoded < 0)
      throw new RuntimeException("got error decoding audio in: " + filename);

    /*
     * Some decoders will consume data in a packet, but will not be able to construct
     * a full video picture yet.  Therefore you should always check if you
     * got a complete picture from the decoder
     */
    if (picture.isComplete())
    {
      IVideoPicture newPic = picture;
      /*
       * If the resampler is not null, that means we didn't get the video in BGR24 format and
       * need to convert it into BGR24 format.
       */
      if (resampler != null)
      {
        // we must resample
        newPic = IVideoPicture.make(resampler.getOutputPixelFormat(), picture.getWidth(), picture.getHeight());
        if (resampler.resample(newPic, picture) < 0)
          throw new RuntimeException("could not resample video from: " + filename);
      }
      if (newPic.getPixelType() != IPixelFormat.Type.BGR24)
        throw new RuntimeException("could not decode video as BGR 24 bit data in: " + filename);

      long delay = millisecondsUntilTimeToDisplay(newPic);
      // if there is no audio stream; go ahead and hold up the main thread.  We'll end
      // up caching fewer video pictures in memory that way.
      try
      {
        if (delay > 0)
          Thread.sleep(delay);
      }
      catch (InterruptedException e)
      {
        return;
      }

      // And finally, convert the picture to an image and display it

      mScreen.setImage(Utils.videoPictureToImage(newPic));
    }
  }
  else if (packet.getStreamIndex() == audioStreamId)
  {
    /*
     * We allocate a set of samples with the same number of channels as the
     * coder tells us is in this buffer.
     * 
     * We also pass in a buffer size (1024 in our example), although Xuggler
     * will probably allocate more space than just the 1024 (it's not important why).
     */
    IAudioSamples samples = IAudioSamples.make(1024, audioCoder.getChannels());

    /*
     * A packet can actually contain multiple sets of samples (or frames of samples
     * in audio-decoding speak).  So, we may need to call decode audio multiple
     * times at different offsets in the packet's data.  We capture that here.
     */
    int offset = 0;

    /*
     * Keep going until we've processed all data
     */
    while(offset < packet.getSize())
    {
      int bytesDecoded = audioCoder.decodeAudio(samples, packet, offset);
      if (bytesDecoded < 0)
        throw new RuntimeException("got error decoding audio in: " + filename);
      offset += bytesDecoded;
      /*
       * Some decoder will consume data in a packet, but will not be able to construct
       * a full set of samples yet.  Therefore you should always check if you
       * got a complete set of samples from the decoder
       */
      if (samples.isComplete())
      {
        // note: this call will block if Java's sound buffers fill up, and we're
        // okay with that.  That's why we have the video "sleeping" occur
        // on another thread.
        playJavaSound(samples);
      }
    }
  }
  else
  {
    /*
     * This packet isn't part of our video stream, so we just silently drop it.
     */
    do {} while(false);
  }

}
/*
 * Technically since we're exiting anyway, these will be cleaned up by 
 * the garbage collector... but because we're nice people and want
 * to be invited places for Christmas, we're going to show how to clean up.
 */
if (videoCoder != null)
{
  videoCoder.close();
  videoCoder = null;
}
if (audioCoder != null)
{
  audioCoder.close();
  audioCoder = null;
}
if (container !=null)
{
  container.close();
  container = null;
}
closeJavaSound();
closeJavaVideo();

} 以下方法有什么作用？

private static long millisecondsUntilTimeToDisplay(IVideoPicture picture)
{
/**
 * We could just display the images as quickly as we decode them, but it turns
 * out we can decode a lot faster than you think.
 * 
 * So instead, the following code does a poor-man's version of trying to
 * match up the frame-rate requested for each IVideoPicture with the system
 * clock time on your computer.
 * 
 * Remember that all Xuggler IAudioSamples and IVideoPicture objects always
 * give timestamps in Microseconds, relative to the first decoded item.  If
 * instead you used the packet timestamps, they can be in different units depending
 * on your IContainer, and IStream and things can get hairy quickly.
 */
long millisecondsToSleep = 0;
if (mFirstVideoTimestampInStream == Global.NO_PTS)
{
  // This is our first time through
  mFirstVideoTimestampInStream = picture.getTimeStamp();
  // get the starting clock time so we can hold up frames
  // until the right time.
  mSystemVideoClockStartTime = System.currentTimeMillis();
  millisecondsToSleep = 0;
} else {
  long systemClockCurrentTime = System.currentTimeMillis();
  long millisecondsClockTimeSinceStartofVideo = systemClockCurrentTime - mSystemVideoClockStartTime;
  // compute how long for this frame since the first frame in the stream.
  // remember that IVideoPicture and IAudioSamples timestamps are always in MICROSECONDS,
  // so we divide by 1000 to get milliseconds.
  long millisecondsStreamTimeSinceStartOfVideo = (picture.getTimeStamp() - mFirstVideoTimestampInStream)/1000;
  final long millisecondsTolerance = 50; // and we give ourselfs 50 ms of tolerance
  millisecondsToSleep = (millisecondsStreamTimeSinceStartOfVideo -
      (millisecondsClockTimeSinceStartofVideo+millisecondsTolerance));
}
return millisecondsToSleep;

/**
 * Opens a Swing window on screen.
 */
 private static void openJavaVideo()
 {
  mScreen = new VideoImage();
 }

 /**
  * Forces the swing thread to terminate; I'm sure there is a right
  * way to do this in swing, but this works too.
  */
  private static void closeJavaVideo()
  {
     System.exit(0);
  }

  private static void openJavaSound(IStreamCoder aAudioCoder) throws     LineUnavailableException
  {
   AudioFormat audioFormat = new AudioFormat(aAudioCoder.getSampleRate(),
     (int)IAudioSamples.findSampleBitDepth(aAudioCoder.getSampleFormat()),
     aAudioCoder.getChannels(),
     true, /* xuggler defaults to signed 16 bit samples */
     false);
   DataLine.Info info = new DataLine.Info(SourceDataLine.class, audioFormat);
   mLine = (SourceDataLine) AudioSystem.getLine(info);
   /**
    * if that succeeded, try opening the line.
    */
    mLine.open(audioFormat);
    /**
     * And if that succeed, start the line.
     */
     mLine.start();   
   }

  private static void playJavaSound(IAudioSamples aSamples)
  {
   /**
    * We're just going to dump all the samples into the line.
    */
    byte[] rawBytes = aSamples.getData().getByteArray(0, aSamples.getSize());
    mLine.write(rawBytes, 0, aSamples.getSize());
  }

  private static void closeJavaSound()
  {
   if (mLine != null)
  {
    /*
     * Wait for the line to finish playing
     */
     mLine.drain();
     /*
      * Close the line.
      */
      mLine.close();
      mLine=null;
     }
    }
 }

Answer 1

伪代码中的粗略算法:

Is this the first frame?
  > Yes, save the frame time and the current time.

  > No, do the following:
    See how much time has passed since the first frame was displayed in System Time
    See the difference in time between the current frame and the first frame

    If there is a discrepancy
      >Return a number of milliseconds to sleep for, else return 0.

那么，你得到的是整体算法:

Decode frame
Check if we need to delay the frame (the method in question)
Delay
Display frame

这样，程序显示帧的速度永远不会超过视频声明的可变帧速率。所讨论的方法保持先前帧时间的状态并计算 hibernate 时间。

编辑:延迟是必需的，因为您可以比视频的帧速率更快地解码帧(快得多!)。假设你有一台相当慢的机器运行这个程序，解码一帧需要 10 毫秒。假设您的视频具有可变帧速率，但大约为每秒 10 帧(或每帧 100 毫秒)。现在，如果您从我们的“整体算法”中取出这一步:

Decode frame (10ms)
Display frame (1ms)
Decode frame (10ms)
Display frame (1ms)

如果发生这种情况，您会发现每 10 毫秒显示 1 帧，这意味着视频将以每秒 100 帧的速度显示，这是错误的!

EDIT2:我猜你想问的是我们为什么不这样做？

Decode frame
Frame Delta = Current Frame Time - Previous Frame Time
Delay (for Delta milliseconds)
Display frame

问题是如果解码或显示帧需要很长时间会发生什么？这会导致帧速率明显低于文件中的帧速率。

相反，该算法将第一帧同步到系统时间，然后进行一些额外的计算:

long systemTimeChange = currentSystemTime - firstFrameSystemTime;
long frameTimeChange = currentFrameTime - firstFrameTime;

// Subtract the time elapsed.
long differenceInChanges = frameTimeChange - systemTimeChange;
if(differenceInChanges > 0) {
  // It was faster to decode than the frame rate!
  Thread.sleep(differenceInChanges);
}