Reprinted PS package format: GB28181 protocol RTP transmission
In the security industry, there is a protocol that must be adapted anyway, because the public security monitoring network uses it, and it is: GB28181. This agreement is mainly formulated by Hikvision, so it will be a little troublesome for other manufacturers except Hikvision to adapt. The former owner of the subject is Haikang. Here is a little analysis of several confusing details of the agreement, and record them here for your own reference in the future and also for your convenience.
1. RTP stream format required by GB28181
First, let's take a look at the PS stream format of the I frame. It should be noted that the PES header must be added before the SPS and PPS. As shown in the figure below, the green part is the H.264 raw stream data we got. It must be split into three segments and the PES header is added to the front. This is not elaborated in the GB28181 standard, and it can be seen only by analyzing the Haikang IPC stream. 
In general, the IDR frame is very large, exceeding the RTP payload length limit (1400 bytes), so the above I frame should be split into several RTP packets and sent multiple times. The structure of the first packet is as shown in the figure above. After the second packet, the structure of RTP is much simpler. It is like this: the 
above mentioned is the case of I frame. Compared with it, the frame format of P/B frame is really too Simple, because it has neither SYS, PSM, nor SPS, PPS: 
P/B frame size generally does not exceed 1400 bytes. If it exceeds 1400 bytes, it must be divided into multiple packets of RTP data for transmission. Two-pack RTP structure:
The first is the RTP header information, which is generally 12 bytes in length
#define RTP_HDR_LEN 12
static int gb28181_make_rtp_header(char *pData, int marker_flag, unsigned short cseq, long long curpts, unsigned int ssrc)
{
bits_buffer_s bitsBuffer;
if (pData == NULL)
return -1;
bitsBuffer.i_size = RTP_HDR_LEN;
bitsBuffer.i_data = 0;
bitsBuffer.i_mask = 0x80;
bitsBuffer.p_data = (unsigned char *)(pData);
memset(bitsBuffer.p_data, 0, RTP_HDR_SIZE);
bits_write(&bitsBuffer, 2, RTP_VERSION); /* rtp version 版本号，固定为2 */
bits_write(&bitsBuffer, 1, 0); /* rtp padding */
bits_write(&bitsBuffer, 1, 0); /* rtp extension */
bits_write(&bitsBuffer, 4, 0); /* rtp CSRC count */
bits_write(&bitsBuffer, 1, (marker_flag)); /* rtp marker 结束标志位，一帧图像的最后一包RTP置1*/
bits_write(&bitsBuffer, 7, 96); /* rtp payload type，96代表PS*/
bits_write(&bitsBuffer, 16, (cseq)); /* rtp sequence */
bits_write(&bitsBuffer, 32, (curpts)); /* rtp timestamp */
bits_write(&bitsBuffer, 32, (ssrc)); /* rtp SSRC */
return 0;
}
Then there is the PSH header, which occupies 14 bytes:
#define PS_HDR_LEN 14
static int gb28181_make_ps_header(char *pData, unsigned long long s64Scr)
{
unsigned long long lScrExt = (s64Scr) % 100;
s64Scr = s64Scr / 100;
bits_buffer_s bitsBuffer;
bitsBuffer.i_size = PS_HDR_LEN;
bitsBuffer.i_data = 0;
bitsBuffer.i_mask = 0x80;
bitsBuffer.p_data = (unsigned char *)(pData);
memset(bitsBuffer.p_data, 0, PS_HDR_LEN);
bits_write(&bitsBuffer, 32, 0x000001BA); /*start codes 起始码*/
bits_write(&bitsBuffer, 2, 1); /*marker bits '01b'*/
bits_write(&bitsBuffer, 3, (s64Scr>>30)&0x07); /*System clock [32..30]*/
bits_write(&bitsBuffer, 1, 1); /*marker bit*/
bits_write(&bitsBuffer, 15, (s64Scr>>15)&0x7FFF); /*System clock [29..15]*/
bits_write(&bitsBuffer, 1, 1); /*marker bit*/
bits_write(&bitsBuffer, 15, s64Scr & 0x7fff); /*System clock [14..0]*/
bits_write(&bitsBuffer, 1, 1); /*marker bit*/
bits_write(&bitsBuffer, 9, 0); /*SCR extension*/
bits_write(&bitsBuffer, 1, 1); /*marker bit*/
bits_write(&bitsBuffer, 22, (255)&0x3fffff); /*bit rate(n units of 50 bytes per second.)*/
bits_write(&bitsBuffer, 2, 3); /*marker bits '11'*/
bits_write(&bitsBuffer, 5, 0x1f); /*reserved(reserved for future use)*/
bits_write(&bitsBuffer, 3, 0); /*stuffing length*/
return 0;
}
Followed by SYS, it contains stream type information, such as audio or video, video encoding format
#define SYS_HDR_LEN 18
static int gb28181_make_sys_header(char *pData)
{
bits_buffer_s bitsBuffer;
bitsBuffer.i_size = SYS_HDR_LEN;
bitsBuffer.i_data = 0;
bitsBuffer.i_mask = 0x80;
bitsBuffer.p_data = (unsigned char *)(pData);
memset(bitsBuffer.p_data, 0, SYS_HDR_LEN);
/*system header*/
bits_write( &bitsBuffer, 32, 0x000001BB); /*start code*/
bits_write( &bitsBuffer, 16, SYS_HDR_LEN-6); /* 减6，是因为start code加上length这两位，占了6个字节*/
bits_write( &bitsBuffer, 1, 1); /*marker_bit*/
bits_write( &bitsBuffer, 22, 50000); /*rate_bound*/
bits_write( &bitsBuffer, 1, 1); /*marker_bit*/
bits_write( &bitsBuffer, 6, 1); /*audio_bound*/
bits_write( &bitsBuffer, 1, 0); /*fixed_flag */
bits_write( &bitsBuffer, 1, 1); /*CSPS_flag */
bits_write( &bitsBuffer, 1, 1); /*system_audio_lock_flag*/
bits_write( &bitsBuffer, 1, 1); /*system_video_lock_flag*/
bits_write( &bitsBuffer, 1, 1); /*marker_bit*/
bits_write( &bitsBuffer, 5, 1); /*video_bound*/
bits_write( &bitsBuffer, 1, 0); /*dif from mpeg1*/
bits_write( &bitsBuffer, 7, 0x7F); /*reserver*/
/*audio stream bound*/
bits_write( &bitsBuffer, 8, 0xC0); /*stream_id 音频的流id*/
bits_write( &bitsBuffer, 2, 3); /*marker_bit */
bits_write( &bitsBuffer, 1, 0); /*PSTD_buffer_bound_scale*/
bits_write( &bitsBuffer, 13, 512); /*PSTD_buffer_size_bound*/
/*video stream bound*/
bits_write( &bitsBuffer, 8, 0xE0); /*stream_id 视频的流id*/
bits_write( &bitsBuffer, 2, 3); /*marker_bit */
bits_write( &bitsBuffer, 1, 1); /*PSTD_buffer_bound_scale*/
bits_write( &bitsBuffer, 13, 2048); /*PSTD_buffer_size_bound*/
return 0;
}
Next is PSM, which records media information, such as the encoding format of audio and video:
static int gb28181_make_psm_header(char *pData)
{
bits_buffer_s bitsBuffer;
bitsBuffer.i_size = PSM_HDR_LEN;
bitsBuffer.i_data = 0;
bitsBuffer.i_mask = 0x80;
bitsBuffer.p_data = (unsigned char *)(pData);
memset(bitsBuffer.p_data, 0, PSM_HDR_LEN);//24Bytes
bits_write(&bitsBuffer, 24,0x000001); /*start code*/
bits_write(&bitsBuffer, 8, 0xBC); /*map stream id*/
bits_write(&bitsBuffer, 16,18); /*program stream map length*/
bits_write(&bitsBuffer, 1, 1); /*current next indicator */
bits_write(&bitsBuffer, 2, 3); /*reserved*/
bits_write(&bitsBuffer, 5, 0); /*program stream map version*/
bits_write(&bitsBuffer, 7, 0x7F); /*reserved */
bits_write(&bitsBuffer, 1, 1); /*marker bit */
bits_write(&bitsBuffer, 16,0); /*programe stream info length*/
bits_write(&bitsBuffer, 16, 8); /*elementary stream map length is*/
/*audio*/
bits_write(&bitsBuffer, 8, 0x90); /*stream_type 音频编码格式G711*/
bits_write(&bitsBuffer, 8, 0xC0); /*elementary_stream_id*/
bits_write(&bitsBuffer, 16, 0); /*elementary_stream_info_length is*/
/*video*/
bits_write(&bitsBuffer, 8, 0x1B); /*stream_type 视频编码格式H.264*/
bits_write(&bitsBuffer, 8, 0xE0); /*elementary_stream_id*/
bits_write(&bitsBuffer, 16, 0); /*elementary_stream_info_length */
/*crc (2e b9 0f 3d)*/
bits_write(&bitsBuffer, 8, 0x45); /*crc (24~31) bits*/
bits_write(&bitsBuffer, 8, 0xBD); /*crc (16~23) bits*/
bits_write(&bitsBuffer, 8, 0xDC); /*crc (8~15) bits*/
bits_write(&bitsBuffer, 8, 0xF4); /*crc (0~7) bits*/
return 0;
}
The last is the PES header, which records the timestamp of the frame. DTS can be left blank. If the input should be consistent with the PTS, and the PTS of the same frame of data should be the same (that is, the PES of SPS, PPS, and IDR should be consistent):
#define PES_HDR_LEN 19
static int gb28181_make_pes_header(char *pData, int stream_id, int payload_len, unsigned long long pts, unsigned long long dts)
{
bits_buffer_s bitsBuffer;
bitsBuffer.i_size = PES_HDR_LEN;
bitsBuffer.i_data = 0;
bitsBuffer.i_mask = 0x80;
bitsBuffer.p_data = (unsigned char *)(pData);
memset(bitsBuffer.p_data, 0, PES_HDR_LEN);
/*system header*/
bits_write( &bitsBuffer, 24,0x000001); /*start code*/
bits_write( &bitsBuffer, 8, (stream_id)); /*streamID*/
bits_write( &bitsBuffer, 16,(payload_len)+13); /*packet_len pes剩余头部以及后面的es长度之和，比如SPS长度+13*/
bits_write( &bitsBuffer, 2, 2 ); /*'10'*/
bits_write( &bitsBuffer, 2, 0 ); /*scrambling_control*/
bits_write( &bitsBuffer, 1, 1 ); /*priority*/
bits_write( &bitsBuffer, 1, 1 ); /*data_alignment_indicator*/
bits_write( &bitsBuffer, 1, 0 ); /*copyright*/
bits_write( &bitsBuffer, 1, 0 ); /*original_or_copy*/
bits_write( &bitsBuffer, 1, 1 ); /*PTS_flag 是否有PTS*/
bits_write( &bitsBuffer, 1, 1 ); /*DTS_flag 是否有DTS信息*/
bits_write( &bitsBuffer, 1, 0 ); /*ESCR_flag*/
bits_write( &bitsBuffer, 1, 0 ); /*ES_rate_flag*/
bits_write( &bitsBuffer, 1, 0 ); /*DSM_trick_mode_flag*/
bits_write( &bitsBuffer, 1, 0 ); /*additional_copy_info_flag*/
bits_write( &bitsBuffer, 1, 0 ); /*PES_CRC_flag*/
bits_write( &bitsBuffer, 1, 0 ); /*PES_extension_flag*/
bits_write( &bitsBuffer, 8, 10); /*header_data_length*/
/*PTS,DTS*/
bits_write( &bitsBuffer, 4, 3 ); /*'0011'*/
bits_write( &bitsBuffer, 3, ((pts)>>30)&0x07 ); /*PTS[32..30]*/
bits_write( &bitsBuffer, 1, 1 );
bits_write( &bitsBuffer, 15,((pts)>>15)&0x7FFF); /*PTS[29..15]*/
bits_write( &bitsBuffer, 1, 1 );
bits_write( &bitsBuffer, 15,(pts)&0x7FFF); /*PTS[14..0]*/
bits_write( &bitsBuffer, 1, 1 );
bits_write( &bitsBuffer, 4, 1 ); /*'0001'*/
bits_write( &bitsBuffer, 3, ((dts)>>30)&0x07 ); /*DTS[32..30]*/
bits_write( &bitsBuffer, 1, 1 );
bits_write( &bitsBuffer, 15,((dts)>>15)&0x7FFF); /*DTS[29..15]*/
bits_write( &bitsBuffer, 1, 1 );
bits_write( &bitsBuffer, 15,(dts)&0x7FFF); /*DTS[14..0]*/
bits_write( &bitsBuffer, 1, 1 );
return 0;
}
Regarding the bit structure of the timestamp, the following figure shows it more clearly: