live555中关于mpeg4的处理

    live555支持mpeg4的ES(Elemental Stream)流，相关类为MPEGVideoStreamFramer、MPEG4ESVideoRTPSink。我想扩展其对avi格式的支持，将avi中的MPEG4数据包解析出来后，交给MPEGVideoStreamFramer进行处理。后来发现，这样根本不行。问题在于，MPEGVideoStreamFramer处理的是严格的MPEG4 ES流。

    先简单的说明一下MPEG4的ES流：

MPEG4 Elemental stream 组成如下：
VOS->VO->VOL->GOV(可选)->VOP
VOS 视觉对像序列     
VO 视觉对像
VOL 视觉对对象层
GOV 视觉对象平面组(VOP组)
VOP 视觉对象平面 

   紧跟着VOP开始的，有一个2bit 的标志，用来表示这个Frame到底是一个 I Frame，P Frame，B Frame抑或是S Frame（GMS-VOP）
标志如下：
00: I Frame
01: P Frame
10: B Frame
11: S Frame


起始符及结束符定义如下:

#define VISUAL_OBJECT_SEQUENCE_START_CODE 0x000001B0
#define VISUAL_OBJECT_SEQUENCE_END_CODE   0x000001B1
#define GROUP_VOP_START_CODE              0x000001B3
#define VISUAL_OBJECT_START_CODE          0x000001B5
#define VOP_START_CODE                    0x000001B6

    用二进制方式打开avi文件，发现只存在vop开始符，说明只存在VOP层次，而不是严格的ES流。可以认为一个VOP对应着一个帧。


    后来发现，live555中实现了另一个类，MPEG4VideoStreamDiscreteFramer， 继承自MPEG4VideoStreamFramer。它可以处理VOS，也可以处理一个个的BOV及VOP，正好可以满足需求。
    看一下MPEG4VideoStreamDiscreteFramer对MPEG4数据的处理

void MPEG4VideoStreamDiscreteFramer
::afterGettingFrame1(unsigned frameSize, unsigned numTruncatedBytes,
                     struct timeval presentationTime,
                     unsigned durationInMicroseconds) {
  // Check that the first 4 bytes are a system code:
  if (frameSize >= 4 && fTo[0] == 0 && fTo[1] == 0 && fTo[2] == 1) {
    fPictureEndMarker = True; // Assume that we have a complete 'picture' here
    unsigned i = 3;


    //
    //视觉对象序列，按照完整的MPEG4 Elemental Stream进行解析
    //
    if (fTo[i] == 0xB0) { // VISUAL_OBJECT_SEQUENCE_START_CODE
      // The next byte is the "profile_and_level_indication":
      if (frameSize >= 5) fProfileAndLevelIndication = fTo[4];


      // The start of this frame - up to the first GROUP_VOP_START_CODE
      // or VOP_START_CODE - is stream configuration information.  Save this:
      for (i = 7; i < frameSize; ++i) {
	if ((fTo[i] == 0xB3 /*GROUP_VOP_START_CODE*/ ||
	     fTo[i] == 0xB6 /*VOP_START_CODE*/)
	    && fTo[i-1] == 1 && fTo[i-2] == 0 && fTo[i-3] == 0) {
	  break; // The configuration information ends here
	}
      }
      fNumConfigBytes = i < frameSize ? i-3 : frameSize;
      delete[] fConfigBytes; fConfigBytes = new unsigned char[fNumConfigBytes];
      for (unsigned j = 0; j < fNumConfigBytes; ++j) fConfigBytes[j] = fTo[j];


      // This information (should) also contain a VOL header, which we need
      // to analyze, to get "vop_time_increment_resolution" (which we need
      // - along with "vop_time_increment" - in order to generate accurate
      // presentation times for "B" frames).
      analyzeVOLHeader();
    }


    if (i < frameSize) {
      u_int8_t nextCode = fTo[i];


      //
      //VOP组
      //
      if (nextCode == 0xB3 /*GROUP_VOP_START_CODE*/) {
	// Skip to the following VOP_START_CODE (if any):
	for (i += 4; i < frameSize; ++i) {
	  if (fTo[i] == 0xB6 /*VOP_START_CODE*/
	      && fTo[i-1] == 1 && fTo[i-2] == 0 && fTo[i-3] == 0) {
	    nextCode = fTo[i];
	    break;
	  }
	}
      }


      //
      //视觉对象平面
      //
      if (nextCode == 0xB6 /*VOP_START_CODE*/ && i+5 < frameSize) {
	++i;


	// Get the "vop_coding_type" from the next byte:
	u_int8_t nextByte = fTo[i++];
	u_int8_t vop_coding_type = nextByte>>6;   //VOP开始符后的2bit，表示帧类型I/P/B/S


	// Next, get the "modulo_time_base" by counting the '1' bits that
	// follow.  We look at the next 32-bits only.
	// This should be enough in most cases.
	u_int32_t next4Bytes
	  = (fTo[i]<<24)|(fTo[i+1]<<16)|(fTo[i+2]<<8)|fTo[i+3];
	i += 4;
	u_int32_t timeInfo = (nextByte<<(32-6))|(next4Bytes>>6);
	unsigned modulo_time_base = 0;
	u_int32_t mask = 0x80000000;
	while ((timeInfo&mask) != 0) {
	  ++modulo_time_base;
	  mask >>= 1;
	}
	mask >>= 2;


	// Then, get the "vop_time_increment".
	unsigned vop_time_increment = 0;
	// First, make sure we have enough bits left for this:
	if ((mask>>(fNumVTIRBits-1)) != 0) {
	  for (unsigned i = 0; i < fNumVTIRBits; ++i) {
	    vop_time_increment |= timeInfo&mask;
	    mask >>= 1;
	  }
	  while (mask != 0) {
	    vop_time_increment >>= 1;
	    mask >>= 1;
	  }
	}


    //
    //若是"B"frame, 需要修正时间时间戳
    //
	// If this is a "B" frame, then we have to tweak "presentationTime":
	if (vop_coding_type == 2/*B*/
	    && (fLastNonBFramePresentationTime.tv_usec > 0 ||
		fLastNonBFramePresentationTime.tv_sec > 0)) {
	  int timeIncrement
	    = fLastNonBFrameVop_time_increment - vop_time_increment;
	  if (timeIncrement<0) timeIncrement += vop_time_increment_resolution;
	  unsigned const MILLION = 1000000;
	  double usIncrement = vop_time_increment_resolution == 0 ? 0.0
	    : ((double)timeIncrement*MILLION)/vop_time_increment_resolution;
	  unsigned secondsToSubtract = (unsigned)(usIncrement/MILLION);
	  unsigned uSecondsToSubtract = ((unsigned)usIncrement)%MILLION;


	  presentationTime = fLastNonBFramePresentationTime;
	  if ((unsigned)presentationTime.tv_usec < uSecondsToSubtract) {
	    presentationTime.tv_usec += MILLION;
	    if (presentationTime.tv_sec > 0) --presentationTime.tv_sec;
	  }
	  presentationTime.tv_usec -= uSecondsToSubtract;
	  if ((unsigned)presentationTime.tv_sec > secondsToSubtract) {
	    presentationTime.tv_sec -= secondsToSubtract;
	  } else {
	    presentationTime.tv_sec = presentationTime.tv_usec = 0;
	  }
	} else {
	  fLastNonBFramePresentationTime = presentationTime;
	  fLastNonBFrameVop_time_increment = vop_time_increment;
	}
      }
    }
  }


  // Complete delivery to the client:
  fFrameSize = frameSize;
  fNumTruncatedBytes = numTruncatedBytes;
  fPresentationTime = presentationTime;
  fDurationInMicroseconds = durationInMicroseconds;
  afterGetting(this);
}

上面的代码，其实只完成一个功能，就是当当前VOP为B帧时，调整时间戳。

最后关注一下，MPEG4 ES流时间戳的处理。 在处理MPEG4 的ES流时，使用MPEG4VideoStreamFramer，作为source。使用分析器MPEG4VideoStreamParser，对完整的MPEG4 Elemental Stream进行分析，主要是解析出其中的时间信息。

void MPEGVideoStreamFramer::continueReadProcessing() {
  unsigned acquiredFrameSize = fParser->parse();
  if (acquiredFrameSize > 0) {
    // We were able to acquire a frame from the input.
    // It has already been copied to the reader's space.
    fFrameSize = acquiredFrameSize;
    fNumTruncatedBytes = fParser->numTruncatedBytes();


    // "fPresentationTime" should have already been computed.


    //
    //根据帧计数及帧率计算帧的持续时间
    //
    // Compute "fDurationInMicroseconds" now:
    fDurationInMicroseconds
      = (fFrameRate == 0.0 || ((int)fPictureCount) < 0) ? 0
      : (unsigned)((fPictureCount*1000000)/fFrameRate);


    fPictureCount = 0;


    // Call our own 'after getting' function.  Because we're not a 'leaf'
    // source, we can call this directly, without risking infinite recursion.
    afterGetting(this);
  } else {
    // We were unable to parse a complete frame from the input, because:
    // - we had to read more data from the source stream, or
    // - the source stream has ended.
  }
}

计算fDurationInMicroseconds需要frame rate参数fFrameRate, 它是通过分析VOL头确定的

void MPEG4VideoStreamParser::analyzeVOLHeader() {
    //
    //从VOL中解析出时间信息
    //
  // Extract timing information (in particular,
  // "vop_time_increment_resolution") from the VOL Header:
...
  do {


      ...


    // Use "vop_time_increment_resolution" as the 'frame rate'
    // (really, 'tick rate'):
    usingSource()->fFrameRate = (double)vop_time_increment_resolution;  //帧率


    return;
  } while (0);

...
}