神剑山庄资源网 Design By www.hcban.com

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()

以上这篇树莓派使用python-librtmp实现rtmp推流h264的方法就是小编分享给大家的全部内容了,希望能给大家一个参考,也希望大家多多支持。

标签:
python,librtmp,rtmp,h264

神剑山庄资源网 Design By www.hcban.com
神剑山庄资源网 免责声明:本站文章均来自网站采集或用户投稿,网站不提供任何软件下载或自行开发的软件! 如有用户或公司发现本站内容信息存在侵权行为,请邮件告知! 858582#qq.com
神剑山庄资源网 Design By www.hcban.com

评论“树莓派使用python-librtmp实现rtmp推流h264的方法”

暂无树莓派使用python-librtmp实现rtmp推流h264的方法的评论...