RTMP 推流增加对H265的支持

RTMP协议本身是不支持H265的。但现在的设备越来越追求更高的压缩比和更高的图形质量。H265相对其他的媒体格式更多受到厂家的重视。rtmp协议要支持H265首先要定义一个ID。按照大家的约定来看，基本使用12(0xc)作为ID. 同时相对H264对NALU的分析要进行改变。并对发送的Metadata数据进行修改。

先看下发送metadata：

int SendVideoSpsPpsVps(RTMP* r, unsigned char* pps, int pps_len, unsigned char* sps, int sps_len, unsigned char* vps,
	int vps_len, uint32_t dts)
{
	char        tBuffer[RTMP_HEAD_SIZE + 1024] = { 0 };
	RTMPPacket* packet = (RTMPPacket*)tBuffer;
	packet->m_body = (char*)packet + RTMP_HEAD_SIZE;
	unsigned char* body = (unsigned char*)packet->m_body;
	// http://ffmpeg.org/doxygen/trunk/hevc_8c_source.html#l00040  hvcc_write 函数
	// 在nginx-rtmp中会跳过48位不去处理 我们在表示后面补0
	// skip tag header and configurationVersion(1 byte)
	int i = 0;
	body[i++] = 0x1C;
	body[i++] = 0x0;
	body[i++] = 0x0;
	body[i++] = 0x0;
	body[i++] = 0x0;
	body[i++] = 0x1;
	// general_profile_idc 8bit
	body[i++] = sps[1];
	// general_profile_compatibility_flags 32 bit
	body[i++] = sps[2];
	body[i++] = sps[3];
	body[i++] = sps[4];
	body[i++] = sps[5];

	// 48 bit NUll nothing deal in rtmp
	body[i++] = sps[6];
	body[i++] = sps[7];
	body[i++] = sps[8];
	body[i++] = sps[9];
	body[i++] = sps[10];
	body[i++] = sps[11];

	// general_level_idc
	body[i++] = sps[12];

	// 48 bit NUll nothing deal in rtmp
	body[i++] = 0;
	body[i++] = 0;
	body[i++] = 0;
	body[i++] = 0;
	body[i++] = 0;
	body[i++] = 0;
	body[i++] = 0;
	body[i++] = 0;
	// bit(16) avgFrameRate;
	/* bit(2) constantFrameRate; */
	/* bit(3) numTemporalLayers; */
	/* bit(1) temporalIdNested; */
	body[i++] = 0x83;

	/* unsigned int(8) numOfArrays; 03 */
	body[i++] = 0x03;
	// vps 32
	body[i++] = 0x20;
	body[i++] = (1 >> 8) & 0xff;
	body[i++] = 1 & 0xff;
	body[i++] = (vps_len >> 8) & 0xff;
	body[i++] = (vps_len) & 0xff;
	memcpy(&body[i], vps, vps_len);
	i += vps_len;

	// sps
	body[i++] = 0x21;  // sps 33
	body[i++] = 0;
	body[i++] = 1;
	body[i++] = (sps_len >> 8) & 0xff;
	body[i++] = sps_len & 0xff;
	memcpy(&body[i], sps, sps_len);
	i += sps_len;

	// pps
	body[i++] = 0x22;  // pps 34
	body[i++] = (1 >> 8) & 0xff;
	body[i++] = 1 & 0xff;
	body[i++] = (pps_len >> 8) & 0xff;
	body[i++] = (pps_len) & 0xff;
	memcpy(&body[i], pps, pps_len);
	i += pps_len;
	packet->m_packetType = RTMP_PACKET_TYPE_VIDEO;
	packet->m_nBodySize = i;
	packet->m_nChannel = 0x04;
	packet->m_nTimeStamp = dts;
	packet->m_hasAbsTimestamp = 0;
	packet->m_headerType = RTMP_PACKET_SIZE_LARGE;
	packet->m_nInfoField2 = r->m_stream_id;
	int nRet = 0;
	if (RTMP_IsConnected(r))
		nRet = RTMP_SendPacket(r, packet, 0);  // 1为放进发送队列,0是不放进发送队列,直接发送
	return nRet;
}

上面中需要注意// bit(16) avgFrameRate;
    /* bit(2) constantFrameRate; */
    /* bit(3) numTemporalLayers; */
    /* bit(1) temporalIdNested; */
    body[i++] = 0x83;

这个地方在一些网站上写的是0，或者不处理，可能造成一些服务器，不工作。

其他，在发送媒体信息的时候需要解释sps。对H265的解释跟H264不一样。

#ifndef h265_decode_info_h__
#define h265_decode_info_h__


#include <iostream>   
#include <sstream>   
//#include <unistd.h>  
#include <stdint.h>
#include <stdio.h>   
#include <string.h>   
#include <stdlib.h>
struct vc_params_t
{
	LONG width, height;
	DWORD profile, level;
	DWORD nal_length_size;
	void clear()
	{
		memset(this, 0, sizeof(*this));
	}
};

class NALBitstream
{
public:
	NALBitstream() : m_data(NULL), m_len(0), m_idx(0), m_bits(0), m_byte(0), m_zeros(0) {};
	NALBitstream(void * data, int len) { Init(data, len); };
	void Init(void * data, int len) { m_data = (LPBYTE)data; m_len = len; m_idx = 0; m_bits = 0; m_byte = 0; m_zeros = 0; };


	BYTE GetBYTE()
	{
		//printf("m_idx=%d,m_len=%d\n", m_idx, m_len);
		if (m_idx >= m_len)
			return 0;
		BYTE b = m_data[m_idx++];

		// to avoid start-code emulation, a byte 0x03 is inserted   
		// after any 00 00 pair. Discard that here.   
		if (b == 0)
		{
			m_zeros++;
			if ((m_idx < m_len) && (m_zeros == 2) && (m_data[m_idx] == 0x03))
			{

				m_idx++;
				m_zeros = 0;
			}
		}

		else
		{
			m_zeros = 0;

		}
		return b;
	};


	UINT32 GetBit()
	{
		
		if (m_bits == 0)
		{
			m_byte = GetBYTE();
			m_bits = 8;

		}
		m_bits--;
		return (m_byte >> m_bits) & 0x1;
	};

	UINT32 GetWord(int bits)
	{

		UINT32 u = 0;
		while (bits > 0)

		{
			u <<= 1;
			u |= GetBit();
			bits--;
		}
		return u;
	};
	UINT32 GetUE()
	{

		// Exp-Golomb entropy coding: leading zeros, then a one, then   
		// the data bits. The number of leading zeros is the number of   
		// data bits, counting up from that number of 1s as the base.   
		// That is, if you see   
		//      0001010   
		// You have three leading zeros, so there are three data bits (010)   
		// counting up from a base of 111: thus 111 + 010 = 1001 = 9   
		int zeros = 0;
		while (m_idx < m_len && GetBit() == 0) zeros++;
		return GetWord(zeros) + ((1 << zeros) - 1);
	};


	INT32 GetSE()
	{

		// same as UE but signed.   
		// basically the unsigned numbers are used as codes to indicate signed numbers in pairs   
		// in increasing value. Thus the encoded values   
		//      0, 1, 2, 3, 4   
		// mean   
		//      0, 1, -1, 2, -2 etc   
		UINT32 UE = GetUE();
		bool positive = UE & 1;
		INT32 SE = (UE + 1) >> 1;
		if (!positive)
		{
			SE = -SE;
		}
		return SE;
	};


private:
	LPBYTE m_data;
	int m_len;
	int m_idx;
	int m_bits;
	BYTE m_byte;
	int m_zeros;
};


bool  ParseSequenceParameterSet(BYTE* data, int size, vc_params_t& params)
{
	if (size < 20)
	{
		return false;
	}

	NALBitstream bs(data, size);

	// seq_parameter_set_rbsp()   
	bs.GetWord(4);// sps_video_parameter_set_id   
	int sps_max_sub_layers_minus1 = bs.GetWord(3); // "The value of sps_max_sub_layers_minus1 shall be in the range of 0 to 6, inclusive."   
	if (sps_max_sub_layers_minus1 > 6)
	{
		return false;
	}
	bs.GetWord(1);// sps_temporal_id_nesting_flag   
				  // profile_tier_level( sps_max_sub_layers_minus1 )   
	{
		bs.GetWord(2);// general_profile_space   
		bs.GetWord(1);// general_tier_flag   
		params.profile = bs.GetWord(5);// general_profile_idc   
		bs.GetWord(32);// general_profile_compatibility_flag[32]   
		bs.GetWord(1);// general_progressive_source_flag   
		bs.GetWord(1);// general_interlaced_source_flag   
		bs.GetWord(1);// general_non_packed_constraint_flag   
		bs.GetWord(1);// general_frame_only_constraint_flag   
		bs.GetWord(44);// general_reserved_zero_44bits   
		params.level = bs.GetWord(8);// general_level_idc   
		unsigned char sub_layer_profile_present_flag[6] = { 0 };
		unsigned char sub_layer_level_present_flag[6] = { 0 };
		for (int i = 0; i < sps_max_sub_layers_minus1; i++)
		{
			sub_layer_profile_present_flag[i] = bs.GetWord(1);
			sub_layer_level_present_flag[i] = bs.GetWord(1);
		}
		if (sps_max_sub_layers_minus1 > 0)
		{
			for (int i = sps_max_sub_layers_minus1; i < 8; i++)
			{
				unsigned char reserved_zero_2bits = bs.GetWord(2);
			}
		}
		for (int i = 0; i < sps_max_sub_layers_minus1; i++)
		{
			if (sub_layer_profile_present_flag[i])
			{
				bs.GetWord(2);// sub_layer_profile_space[i]   
				bs.GetWord(1);// sub_layer_tier_flag[i]   
				bs.GetWord(5);// sub_layer_profile_idc[i]   
				bs.GetWord(32);// sub_layer_profile_compatibility_flag[i][32]   
				bs.GetWord(1);// sub_layer_progressive_source_flag[i]   
				bs.GetWord(1);// sub_layer_interlaced_source_flag[i]   
				bs.GetWord(1);// sub_layer_non_packed_constraint_flag[i]   
				bs.GetWord(1);// sub_layer_frame_only_constraint_flag[i]   
				bs.GetWord(44);// sub_layer_reserved_zero_44bits[i]   
			}
			if (sub_layer_level_present_flag[i])
			{
				bs.GetWord(8);// sub_layer_level_idc[i]   
			}
		}
	}
	unsigned long sps_seq_parameter_set_id = bs.GetUE(); // "The  value  of sps_seq_parameter_set_id shall be in the range of 0 to 15, inclusive."   
	/*if (sps_seq_parameter_set_id > 15)
	{
		printf("enter2\r\n");
		return false;
	}*/
	unsigned long chroma_format_idc = bs.GetUE(); // "The value of chroma_format_idc shall be in the range of 0 to 3, inclusive."   
	/*if (sps_seq_parameter_set_id > 3)
	{
		printf("enter3\r\n");
		return false;
	}*/
	if (chroma_format_idc == 3)
	{
		bs.GetWord(1);// separate_colour_plane_flag   
	}
	params.width = bs.GetUE(); // pic_width_in_luma_samples   
	params.height = bs.GetUE(); // pic_height_in_luma_samples   
	if (bs.GetWord(1))
	{// conformance_window_flag   
		bs.GetUE();  // conf_win_left_offset   
		bs.GetUE();  // conf_win_right_offset   
		bs.GetUE();  // conf_win_top_offset   
		bs.GetUE();  // conf_win_bottom_offset   
	}
	unsigned long bit_depth_luma_minus8 = bs.GetUE();
	unsigned long bit_depth_chroma_minus8 = bs.GetUE();
	/*if (bit_depth_luma_minus8 != bit_depth_chroma_minus8)
	{
		printf("enter4\r\n");
		return false;
	}*/
	//...   


	return true;
}

#endif // h265_decode_info_h__

这样可以发送根据媒体格式进行头信息填写了。

if(lpMetaData == NULL)
	{
		return -1;
	}
	char buffer[1024] = {0};
	char *body = buffer+RTMP_MAX_HEADER_SIZE;

	char * p = (char *)body;  
	p = put_byte(p, AMF_STRING );
	p = put_amf_string(p , "@setDataFrame" );

	p = put_byte( p, AMF_STRING );
	p = put_amf_string( p, "onMetaData" );

	p = put_byte(p, AMF_OBJECT );  
	p = put_amf_string( p, "copyright" );  
	p = put_byte(p, AMF_STRING );  
	p = put_amf_string( p, "CarEyeRTMP" );
	
	if (type == 1)
	{
		p = put_amf_string(p, "width");
		p = put_amf_double(p, lpMetaData->Width);

		p = put_amf_string(p, "height");
		p = put_amf_double(p, lpMetaData->Height);

		p = put_amf_string(p, "framerate");
		p = put_amf_double(p, lpMetaData->FrameRate);

		p = put_amf_string(p, "videocodecid");
		if (lpMetaData->VCodec == CAREYE_VCODE_H264)
		{
			p = put_amf_double(p, FLV_CODECID_H264);
		}
		else
		{
			p = put_amf_double(p, FLV_CODECID_H265);
		}
	}

	p =put_amf_string( p, "audiosamplerate");
	p =put_amf_double( p, lpMetaData->SampleRate);

	p =put_amf_string( p, "audiocodecid");
	p =put_amf_double( p, 10);

	p =put_amf_string( p, "" );
	p =put_byte( p, AMF_OBJECT_END  );

car-eye RTMP推流是将GB28181或者GT1078协议的数据的音视频数据推送到RTMP拉流服务器。以实现客户端对RTMP,http,websocket，HLS等多种方式的拉取和播放。

car-eye流媒体服务器实现了对监控和车载移动设备多种场景的支持。相关的开源源码地址：https://github.com/Car-eye-team/

https://gitee.com/careye_open_source_platform_group

本文章参考：https://blog.csdn.net/qq_33795447/article/details/89457581