SkeyePlayer源码解析系列之支持H265

2023-04-21 10:15:48 浏览数 (2)

近期SkeyePlayer(windows)更新已全面支持H265的RTSP流的解码播放,这里就支持H265过程做简要介绍;

一、 libSkeyeRTSPClient库已支持H265视频源的RTSP流的拉取和解析

二、H265头解析

代码语言:txt复制
H265和H264类似,不过其NAL type格式更多样化,除了SPS,PPS之外,还增加了VPS,下面就针对H265帧nal 头做简单分析;
代码语言:txt复制
首先,看X265源码中的H265nal头格式定义:
代码语言:txt复制
//H265 NAL type
//this enum have been defined in x265.h
typedef enum tagH265NalUnitType
{
	NAL_UNIT_CODED_SLICE_TRAIL_N = 0,   // 0  
	NAL_UNIT_CODED_SLICE_TRAIL_R,   // 1  

	NAL_UNIT_CODED_SLICE_TSA_N,     // 2  
	NAL_UNIT_CODED_SLICE_TLA,       // 3   // Current name in the spec: TSA_R  

	NAL_UNIT_CODED_SLICE_STSA_N,    // 4  
	NAL_UNIT_CODED_SLICE_STSA_R,    // 5  

	NAL_UNIT_CODED_SLICE_RADL_N,    // 6  
	NAL_UNIT_CODED_SLICE_DLP,       // 7 // Current name in the spec: RADL_R  

	NAL_UNIT_CODED_SLICE_RASL_N,    // 8  
	NAL_UNIT_CODED_SLICE_TFD,       // 9 // Current name in the spec: RASL_R  

	NAL_UNIT_RESERVED_10,
	NAL_UNIT_RESERVED_11,
	NAL_UNIT_RESERVED_12,
	NAL_UNIT_RESERVED_13,
	NAL_UNIT_RESERVED_14,
	NAL_UNIT_RESERVED_15,			
	NAL_UNIT_CODED_SLICE_BLA,       // 16   // Current name in the spec: BLA_W_LP  
	NAL_UNIT_CODED_SLICE_BLANT,     // 17   // Current name in the spec: BLA_W_DLP  
	NAL_UNIT_CODED_SLICE_BLA_N_LP,  // 18  
	NAL_UNIT_CODED_SLICE_IDR,       // 19  // Current name in the spec: IDR_W_DLP  
	NAL_UNIT_CODED_SLICE_IDR_N_LP,  // 20  
	NAL_UNIT_CODED_SLICE_CRA,       // 21  
	NAL_UNIT_RESERVED_22,
	NAL_UNIT_RESERVED_23,

	NAL_UNIT_RESERVED_24,
	NAL_UNIT_RESERVED_25,
	NAL_UNIT_RESERVED_26,
	NAL_UNIT_RESERVED_27,
	NAL_UNIT_RESERVED_28,
	NAL_UNIT_RESERVED_29,
	NAL_UNIT_RESERVED_30,
	NAL_UNIT_RESERVED_31,

	NAL_UNIT_VPS,                   // 32  
	NAL_UNIT_SPS,                   // 33  
	NAL_UNIT_PPS,                   // 34  
	NAL_UNIT_ACCESS_UNIT_DELIMITER, // 35  
	NAL_UNIT_EOS,                   // 36  
	NAL_UNIT_EOB,                   // 37  
	NAL_UNIT_FILLER_DATA,           // 38  
	NAL_UNIT_SEI,                   // 39 Prefix SEI  
	NAL_UNIT_SEI_SUFFIX,            // 40 Suffix SEI  
	NAL_UNIT_RESERVED_41,
	NAL_UNIT_RESERVED_42,
	NAL_UNIT_RESERVED_43,
	NAL_UNIT_RESERVED_44,
	NAL_UNIT_RESERVED_45,
	NAL_UNIT_RESERVED_46,
	NAL_UNIT_RESERVED_47,
	NAL_UNIT_UNSPECIFIED_48,
	NAL_UNIT_UNSPECIFIED_49,
	NAL_UNIT_UNSPECIFIED_50,
	NAL_UNIT_UNSPECIFIED_51,
	NAL_UNIT_UNSPECIFIED_52,
	NAL_UNIT_UNSPECIFIED_53,
	NAL_UNIT_UNSPECIFIED_54,
	NAL_UNIT_UNSPECIFIED_55,
	NAL_UNIT_UNSPECIFIED_56,
	NAL_UNIT_UNSPECIFIED_57,
	NAL_UNIT_UNSPECIFIED_58,
	NAL_UNIT_UNSPECIFIED_59,
	NAL_UNIT_UNSPECIFIED_60,
	NAL_UNIT_UNSPECIFIED_61,
	NAL_UNIT_UNSPECIFIED_62,
	NAL_UNIT_UNSPECIFIED_63,
	NAL_UNIT_INVALID,
}H265NalUnitType;
#endif

我们可以看到其中VPS, SPS和PPS的定义:

代码语言:txt复制
NAL_UNIT_VPS,                   // 32  
代码语言:txt复制
NAL_UNIT_SPS,                   // 33  
代码语言:txt复制
NAL_UNIT_PPS,                   // 34 

同样,我们也很容易知道P帧NAL type定义是0-9, I帧定义是16-21;可见H265的NAL type定义比H264要多样化,判断也不限制于一种类型;

同时,测试发现,实际H265帧数据中的VPS=0x40 , SPS=0x42, PPS=0x44, 通过换算,我们不难得出:

NALtype*2 = 实际的流中的NaLType;

具体解析过程如下:

代码语言:txt复制
 //输入的pbuf必须包含start code(00 00 00 01)
int GetH265VPSandSPSandPPS(char *pbuf, int bufsize, char *_vps, int *_vpslen, char *_sps, int *_spslen, char *_pps, int *_ppslen)
{
	char vps[512]={0}, sps[512] = {0}, pps[128] = {0};
	int vpslen=0, spslen=0, ppslen=0, i=0, iStartPos=0, ret=-1;
	int iFoundVPS=0, iFoundSPS=0, iFoundPPS=0, iFoundSEI=0;
	if (NULL == pbuf || bufsize<4)	return -1;

	for (i=0; i<bufsize; i  )
	{
		if ( (unsigned char)pbuf[i] == 0x00 && (unsigned char)pbuf[i 1] == 0x00 && 
			 (unsigned char)pbuf[i 2] == 0x00 && (unsigned char)pbuf[i 3] == 0x01 )
		{
			printf("0x%Xn", (unsigned char)pbuf[i 4]);
			switch ((unsigned char)pbuf[i 4])
			{
			case 0x40:		//VPS
				{
					iFoundVPS = 1;
					iStartPos = i 4;
				}
				break;
			case 0x42:		//SPS
				{
					if (iFoundVPS == 0x01 && i>4)
					{
						vpslen = i-iStartPos;
						if (vpslen>256)	return -1;          //vps长度超出范围
						memset(vps, 0x00, sizeof(vps));
						memcpy(vps, pbuf iStartPos, vpslen);
					}

					iStartPos = i 4;
					iFoundSPS = 1;
				}
				break;
			case 0x44:		//PPS
				{
					if (iFoundSPS == 0x01 && i>4)
					{
						spslen = i-iStartPos;
						if (spslen>256)	return -1;
						memset(sps, 0x0, sizeof(sps));
						memcpy(sps, pbuf iStartPos,  spslen);
					}

					iStartPos = i 4;
					iFoundPPS = 1;
				}
				break;
			case 0x4E:		//Prefix SEI  
			case 0x50:		//Suffix SEI 
			case 0x20:		//I frame 16
			case 0x22:		//I frame 17
			case 0x24:		//I frame 18
			case 0x26:		//I frame 19
			case 0x28:		//I frame 20
			case 0x2A:		//I frame 21(acturally we should find naltype 16-21)
				{
					if (iFoundPPS == 0x01 && i>4)
					{
						ppslen = i-iStartPos;
						if (ppslen>256)	return -1;
						memset(pps, 0x0, sizeof(pps));
						memcpy(pps, pbuf iStartPos,  ppslen);
					}
					iStartPos = i 4;
					iFoundSEI = 1;
				}
				break;
			default:
				break;
			}
		}

		if (iFoundSEI == 0x01)		break;
	}
	if (iFoundVPS == 0x01)
    {
        if (vpslen < 1)
        {
            if (bufsize < sizeof(vps))
            {
                vpslen = bufsize-4;
                memset(vps, 0x00, sizeof(vps));
                memcpy(vps, pbuf 4, vpslen);
            }
        }

        if (vpslen > 0)
        {
            if (NULL != _vps)   memcpy(_vps, vps, vpslen);
            if (NULL != _vpslen)    *_vpslen = vpslen;
        }

        ret = 0;
    }
	if (iFoundSPS == 0x01)
    {
        if (spslen < 1)
        {
            if (bufsize < sizeof(sps))
            {
                spslen = bufsize-4;
                memset(sps, 0x00, sizeof(sps));
                memcpy(sps, pbuf 4, spslen);
            }
        }

        if (spslen > 0)
        {
            if (NULL != _sps)   memcpy(_sps, sps, spslen);
            if (NULL != _spslen)    *_spslen = spslen;
        }

        ret = 0;
    }

    if (iFoundPPS == 0x01)
    {
        if (ppslen < 1)
        {
            if (bufsize < sizeof(pps))
            {
                ppslen = bufsize-4;
                memset(pps, 0x00, sizeof(pps));
				memcpy(pps, pbuf 4, ppslen);	//pps
            }
        }
        if (ppslen > 0)
        {
            if (NULL != _pps)   memcpy(_pps, pps, ppslen);
            if (NULL != _ppslen)    *_ppslen = ppslen;
        }
        ret = 0;
    }

    return ret;
}

三、 解码器需支持H265

代码语言:txt复制
解码器直接使用最新的FFMPEG库即支持H265解码,且软解效率还可以,大家如果不知道怎么用,可以去看看ffplay的源码,这里不做过多赘述;这里就SkeyePlayer调用遇到的问题做简单说明:
代码语言:txt复制
1> 旧版的ffmpeg以及live555等对H265的定义是对“H265”子串做的字串格式组合,而新版的FFMPEG使用的自定义的顺序定义的枚举类型,所以在使用过程中可能出现对应不上的情况,比如,在libSkeyeRTSPClient库中对H265的定义为:#define Skeye_SDK_VIDEO_CODEC_H265	0x48323635	/* 1211250229 */

而FFMPEG中定义H265(HEVC)格式为174,SkeyePlayer中进行格式统一代码如下:

代码语言:txt复制
	//H265 codecID改成FFMPEG新版的
	int nCodec = (_frameinfo->codec == Skeye_SDK_VIDEO_CODEC_H265) ? 174 : _frameinfo->codec;

2> SkeyePlayer中之前对关键帧帧解码失败的处理是将以该I帧为关键帧为依托的所有P帧丢弃,当然这从某种程度上是可以避免花屏的,但是测试解码H265时发现,H265的第一个I帧会经常解码失败,经调试发现其实是FFNPEG的解码函数返回没有解码完成的结果被程序判断为解码失败,而这个时候应该不做任何处理等下一次返回的时候就能获取到正确的返回结果了,SkeyePlayer处理如下:

代码语言:txt复制
			nRet = FFD_DecodeVideo3(pDecoderObj->ffDecoder, pbuf, frameinfo.length, pThread->yuvFrame[pThread->decodeYuvIdx].pYuvBuf, frameinfo.width, frameinfo.height, lTimestamp, lTimestamp);
				if (0 != nRet)
				{
					if(nRet == -4)//-4表示为当前帧尚未解码完成,不作为错误判断
					{
							_TRACE("视频帧解码尚未完成[%d]... framesize:%d   X X X X X X X X X Xn", nRet, frameinfo.length, 
							(unsigned char)pbuf[0], (unsigned char)pbuf[1], (unsigned char)pbuf[2], (unsigned char)pbuf[3], (unsigned char)pbuf[4],
							(unsigned char)pbuf[5], (unsigned char)pbuf[6], (unsigned char)pbuf[7], (unsigned char)pbuf[8], (unsigned char)pbuf[9]);
					}
					else
					{
						_TRACE("视频帧解解码失败[%d]... framesize:%d   X X X X X X X X X Xn", nRet, frameinfo.length, 
							(unsigned char)pbuf[0], (unsigned char)pbuf[1], (unsigned char)pbuf[2], (unsigned char)pbuf[3], (unsigned char)pbuf[4],
							(unsigned char)pbuf[5], (unsigned char)pbuf[6], (unsigned char)pbuf[7], (unsigned char)pbuf[8], (unsigned char)pbuf[9]);

						if (frameinfo.type == Skeye_SDK_VIDEO_FRAME_I)		//关键帧
						{
							_TRACE("[ch%d]当前关键帧解码失败...n", pThread->channelId);
	#ifdef _DEBUG
							FILE *f = fopen("keyframe.txt", "wb");
							if (NULL != f)
							{
								fwrite(pbuf, 1, frameinfo.length, f);
								fclose(f);
							}
	#endif
						}
						else
						{
	#ifdef _DEBUG
							FILE *f = fopen("pframe.txt", "wb");
							if (NULL != f)
							{
								fwrite(pbuf, 1, frameinfo.length, f);
								fclose(f);
							}
	#endif
						}
						pThread->findKeyframe = 0x01;
					}
代码语言:txt复制
经测试,经过如上代码 修改就能有效的避免的解码H265视频的时候开头帧总是会局部卡帧或者花屏的情况出现。

四、H265格式视频写MP4

这里接着之前SkeyePlayer系列的写MP4篇讲,将H265封装MP4;

1> 解析H265的头,或者VPS,SPS和PPS

代码语言:txt复制
	从H265帧中取出NAL头在上文已经作过讲解这里就不做过多赘述;

2> 写入VPS, SPS和PPS 等关键解码信息

代码语言:txt复制
bool SkeyeMP4Writer::WriteH265VPSSPSandPPS(unsigned char*vps, int vpslen, unsigned char*sps, int spslen,
	unsigned char*pps, int ppslen, int width, int height)
{
	if (m_nCreateFileFlag&ZOUTFILE_FLAG_VIDEO)
	{
		m_videtrackid = gf_isom_new_track(p_file, 0, GF_ISOM_MEDIA_VISUAL, 1000);
		gf_isom_set_track_enabled(p_file, m_videtrackid, 1);
	}
	else
	{
		return false;
	}
	p_videosample = gf_isom_sample_new();
	p_videosample->data = (char*)malloc(1024 * 1024);

	p_hevc_config = gf_odf_hevc_cfg_new();
	p_hevc_config->nal_unit_size = 32 / 8;

	//gf_isom_avc_config_new(p_file, m_videtrackid, p_config, NULL, NULL, &i_videodescidx);
	//gf_isom_set_visual_info(p_file, m_videtrackid, i_videodescidx, width, height);
	//初始化配置
	gf_isom_hevc_config_new(p_file, m_videtrackid, p_hevc_config, NULL, NULL, &i_videodescidx);
	gf_isom_set_nalu_extract_mode(p_file, m_videtrackid, GF_ISOM_NALU_EXTRACT_INSPECT);
	gf_isom_set_cts_packing(p_file, m_videtrackid, GF_TRUE);

	HEVCState hevc = { 0 };
	m_slotsps = { 0 };
	m_slotpps = { 0 };
	m_slotvps = { 0 };
	m_spss = { 0 };
	m_ppss = { 0 };
	m_vpss = { 0 };

	p_hevc_config->configurationVersion = 1;

	//Config vps
	int idx = gf_media_hevc_read_vps((char*)vps, vpslen, &hevc);
	hevc.vps[idx].crc = gf_crc_32((char*)vps, vpslen);
	p_hevc_config->avgFrameRate = hevc.vps[idx].rates[0].avg_pic_rate;
	p_hevc_config->constantFrameRate = hevc.vps[idx].rates[0].constand_pic_rate_idc;
	p_hevc_config->numTemporalLayers = hevc.vps[idx].max_sub_layers;
	p_hevc_config->temporalIdNested = hevc.vps[idx].temporal_id_nesting;

	m_vpss.nalus = gf_list_new();
	gf_list_add(p_hevc_config->param_array, &m_vpss);
	m_vpss.array_completeness = 1;
	m_vpss.type = GF_HEVC_NALU_VID_PARAM;// naltype = VPS
	m_slotvps.id = idx;
	m_slotvps.size = vpslen;
	m_slotvps.data = (char*)malloc(vpslen);
	memcpy(m_slotvps.data, vps, vpslen);
	gf_list_add(m_vpss.nalus, &m_slotvps);

	//Config sps
	idx = gf_media_hevc_read_sps((char*)sps, spslen, &hevc);
	hevc.sps[idx].crc = gf_crc_32((char*)sps, spslen);
	p_hevc_config->profile_space = hevc.sps[idx].ptl.profile_space;
	p_hevc_config->tier_flag = hevc.sps[idx].ptl.tier_flag;
	p_hevc_config->profile_idc = hevc.sps[idx].ptl.profile_idc;

	m_spss.nalus = gf_list_new();
	gf_list_add(p_hevc_config->param_array, &m_spss);
	m_spss.array_completeness = 1;
	m_spss.type = GF_HEVC_NALU_SEQ_PARAM;// naltype = SPS
	m_slotsps.id = idx;
	m_slotsps.size = spslen;
	m_slotsps.data = (char*)malloc(spslen);
	memcpy(m_slotsps.data, sps, spslen);
	gf_list_add(m_spss.nalus, &m_slotsps);
	int act_width = hevc.sps[idx].width;
	int act_height = hevc.sps[idx].height;

	//Config pps
	idx = gf_media_hevc_read_pps((char*)pps, ppslen, &hevc);
	hevc.pps[idx].crc = gf_crc_32((char*)pps, ppslen);

	m_ppss.nalus = gf_list_new();
	gf_list_add(p_hevc_config->param_array, &m_ppss);
	m_ppss.array_completeness = 1;
	m_ppss.type = GF_HEVC_NALU_PIC_PARAM;// naltype = PPS
	m_slotpps.id = idx;
	m_slotpps.size = ppslen;
	m_slotpps.data = (char*)malloc(ppslen);
	memcpy(m_slotpps.data, pps, ppslen);
	gf_list_add(m_ppss.nalus, &m_slotpps);

	gf_isom_set_visual_info(p_file, m_videtrackid, i_videodescidx, act_width, act_height);
	gf_isom_hevc_config_update(p_file, m_videtrackid, 1, p_hevc_config);

	//销毁申请的内存资源
	gf_list_del(m_vpss.nalus);
	gf_list_del(m_spss.nalus);
	gf_list_del(m_ppss.nalus);
	free(m_slotvps.data);
	free(m_slotsps.data);
	free(m_slotpps.data);
	p_hevc_config->param_array = NULL;
	return true;
}

结合写MP4篇我们不难看出,MP4BOX对H265专门封装了个结构函数gf_isom_hevc_config_new()用以对H265参数的设置,设置方法和H264相似,不过对H265处理更加细致,MP4BOX将VPS,SPS,PPS的各个参数拆分出来进行赋值,通过gf_isom_hevc_config_update写入解码参数信息。

注意: H265写MP4尚未加入最新的SkeyePlayer源码,这里是独家首发哦--!(主要是忘了,会在近期加入)

0 人点赞