树莓派使用python-librtmp实现rtmp推流h264的方法

目的是能使用Python进行rtmp推流，方便在h264帧里加入弹幕等操作。

librtmp使用的是0.3.0，使用树莓派noir官方摄像头适配的。

通过wireshark抓ffmpeg的包一点点改动，最终可以在red5和斗鱼上推流了。

没怎么写过python，有不恰当的地方请包涵。

上代码：

# -- coding: utf-8 --
# http://blog.csdn.net/luhanglei
import picamera
import time
import traceback
import ctypes
from librtmp import *

global meta_packet
global start_time

class Writer(): # camera可以通过一个类文件的对象来输出，实现write方法即可
 conn = None # rtmp连接
 sps = None # 记录sps帧，发过以后就不需要再发了（抓包看到ffmpeg是这样的）
 pps = None # 同上
 sps_len = 0 # 同上
 pps_len = 0 # 同上

time_stamp = 0

def __init__(self, conn):
   self.conn = conn

def write(self, data):
   try:
     # 寻找h264帧间隔符
     indexs = []
     index = 0
     data_len = len(data)
     while index < data_len - 3:
       if ord(data[index]) == 0x00 and ord(data[index + 1]) == 0x00 and ord(
           data[index + 2]) == 0x00 and ord(data[index + 3]) == 0x01:
         indexs.append(index)
         index = index + 3
       index = index + 1
     # 寻找h264帧间隔符 完成
     # 通过间隔符个数确定类型，树莓派摄像头的第一帧是sps+pps同时发的
     if len(indexs) == 1: # 非sps pps帧
       buf = data[4: len(data)] # 裁掉原来的头（00 00 00 01），把帧内容拿出来
       buf_len = len(buf)
       type = ord(buf[0]) & 0x1f
       if type == 0x05: # 关键帧，根据wire shark抓包结果，需要拼装sps pps 帧内容 三部分，长度都用4个字节表示
         body0 = 0x17
         data_body_array = [bytes(bytearray(
           [body0, 0x01, 0x00, 0x00, 0x00, (self.sps_len >> 24) & 0xff, (self.sps_len >> 16) & 0xff,
            (self.sps_len >> 8) & 0xff,
            self.sps_len & 0xff])), self.sps,
           bytes(bytearray(
             [(self.pps_len >> 24) & 0xff, (self.pps_len >> 16) & 0xff, (self.pps_len >> 8) & 0xff,
              self.pps_len & 0xff])),
           self.pps,
           bytes(bytearray(
             [(buf_len >> 24) & 0xff, (buf_len >> 16) & 0xff, (buf_len >> 8) & 0xff, (buf_len) & 0xff])),
           buf
         ]
         mbody = ''.join(data_body_array)
         time_stamp = 0 # 第一次发出的时候，发时间戳0，此后发真时间戳
         if self.time_stamp != 0:
           time_stamp = int((time.time() - start_time) * 1000)
         packet_body = RTMPPacket(type=PACKET_TYPE_VIDEO, format=PACKET_SIZE_LARGE, channel=0x06,
                      timestamp=time_stamp, body=mbody)
         packet_body.packet.m_nInfoField2 = 1
       else: # 非关键帧
         body0 = 0x27
         data_body_array = [bytes(bytearray(
           [body0, 0x01, 0x00, 0x00, 0x00, (buf_len >> 24) & 0xff, (buf_len >> 16) & 0xff,
            (buf_len >> 8) & 0xff,
            (buf_len) & 0xff])), buf]
         mbody = ''.join(data_body_array)
         # if (self.time_stamp == 0):
         self.time_stamp = int((time.time() - start_time) * 1000)
         packet_body = RTMPPacket(type=PACKET_TYPE_VIDEO, format=PACKET_SIZE_MEDIUM, channel=0x06,
                      timestamp=self.time_stamp, body=mbody)
       self.conn.send_packet(packet_body)
     elif len(indexs) == 2: # sps pps帧
       if self.sps is not None:
         return
       data_body_array = [bytes(bytearray([0x17, 0x00, 0x00, 0x00, 0x00, 0x01]))]
       sps = data[indexs[0] + 4: indexs[1]]
       sps_len = len(sps)
       pps = data[indexs[1] + 4: len(data)]
       pps_len = len(pps)
       self.sps = sps
       self.sps_len = sps_len
       self.pps = pps
       self.pps_len = pps_len
       data_body_array.append(sps[1:4])
       data_body_array.append(bytes(bytearray([0xff, 0xe1, (sps_len >> 8) & 0xff, sps_len & 0xff])))
       data_body_array.append(sps)
       data_body_array.append(bytes(bytearray([0x01, (pps_len >> 8) & 0xff, pps_len & 0xff])))
       data_body_array.append(pps)
       data_body = ''.join(data_body_array)
       body_packet = RTMPPacket(type=PACKET_TYPE_VIDEO, format=PACKET_SIZE_LARGE, channel=0x06,
                    timestamp=0, body=data_body)
       body_packet.packet.m_nInfoField2 = 1

self.conn.send_packet(meta_packet, queue=True)
       self.conn.send_packet(body_packet, queue=True)
   except Exception, e:
     traceback.print_exc()

def flush(self):
   pass

def get_property_string(string): # 返回两字节string长度及string
 length = len(string)
 return ''.join([chr((length >> 8) & 0xff), chr(length & 0xff), string])

def get_meta_string(string): # 按照meta packet要求格式返回bytes,带02前缀
 return ''.join([chr(0x02), get_property_string(string)])

def get_meta_double(db):
 nums = [0x00]
 fp = ctypes.pointer(ctypes.c_double(db))
 cp = ctypes.cast(fp, ctypes.POINTER(ctypes.c_longlong))
 for i in range(7, -1, -1):
   nums.append((cp.contents.value >> (i * 8)) & 0xff)
 return ''.join(bytes(bytearray(nums)))

def get_meta_boolean(isTrue):
 nums = [0x01]
 if (isTrue):
   nums.append(0x01)
 else:
   nums.append(0x00)
 return ''.join(bytes(bytearray(nums)))

conn = RTMP(
 'rtmp://192.168.199.154/oflaDemo/test', # 推流地址
 live=True)
librtmp.RTMP_EnableWrite(conn.rtmp)
conn.connect()
start_time = time.time()
# 拼装视频格式的数据包
meta_body_array = [get_meta_string('@setDataFrame'), get_meta_string('onMetaData'),
         bytes(bytearray([0x08, 0x00, 0x00, 0x00, 0x06])), # 两个字符串和ECMA array头，共计6个元素,注释掉了音频相关数据
         get_property_string('width'), get_meta_double(640.0),
         get_property_string('height'), get_meta_double(480.0),
         get_property_string('videodatarate'), get_meta_double(0.0),
         get_property_string('framerate'), get_meta_double(25.0),
         get_property_string('videocodecid'), get_meta_double(7.0),
         # get_property_string('audiodatarate'), get_meta_double(125.0),
         # get_property_string('audiosamplerate'), get_meta_double(44100.0),
         # get_property_string('audiosamplesize'), get_meta_double(16.0),
         # get_property_string('stereo'), get_meta_boolean(True),
         # get_property_string('audiocodecid'), get_meta_double(10.0),
         get_property_string('encoder'), get_meta_string('Lavf57.56.101'),
         bytes(bytearray([0x00, 0x00, 0x09]))
         ]
meta_body = ''.join(meta_body_array)
print meta_body.encode('hex')
meta_packet = RTMPPacket(type=PACKET_TYPE_INFO, format=PACKET_SIZE_LARGE, channel=0x04,
            timestamp=0, body=meta_body)
meta_packet.packet.m_nInfoField2 = 1 # 修改stream id
stream = conn.create_stream(writeable=True)
with picamera.PiCamera() as camera:
 camera.start_preview()
 time.sleep(2)
 camera.start_recording(Writer(conn), format='h264', resize=(640, 480), intra_period=25,
             quality=25) # 开始录制，数据输出到Writer的对象里
 while True:#永远不停止
   time.sleep(60)
 camera.stop_recording()
 camera.stop_preview()

来源：https://blog.csdn.net/luhanglei/article/details/78403900