在 python opencv 中通过网络发送实时视频帧

Question

我正在尝试将我用相机捕捉到的实时视频帧发送到服务器并进行处理。我使用 opencv 进行图像处理，使用 python 作为语言。这是我的代码

client_cv.py

import cv2
import numpy as np
import socket
import sys
import pickle
cap=cv2.VideoCapture(0)
clientsocket=socket.socket(socket.AF_INET,socket.SOCK_STREAM)
clientsocket.connect(('localhost',8089))
while True:
    ret,frame=cap.read()
    print sys.getsizeof(frame)
    print frame
    clientsocket.send(pickle.dumps(frame))

server_cv.py

import socket
import sys
import cv2
import pickle
import numpy as np
HOST=''
PORT=8089

s=socket.socket(socket.AF_INET,socket.SOCK_STREAM)
print 'Socket created'

s.bind((HOST,PORT))
print 'Socket bind complete'
s.listen(10)
print 'Socket now listening'

conn,addr=s.accept()

while True:
    data=conn.recv(80)
    print sys.getsizeof(data)
    frame=pickle.loads(data)
    print frame
    cv2.imshow('frame',frame)

这段代码给了我文件结束错误，这是合乎逻辑的，因为数据总是不断地进入服务器，而 pickle 不知道何时完成。我在互联网上的搜索让我使用了泡菜，但到目前为止它不起作用。

注意：我将 conn.recv 设置为 80，因为这是我说 print sys.getsizeof(frame) 时得到的数字。

Answer 1

几件事：

• 请使用sendall 而不是send，因为您不能保证一次性发送所有内容
• pickle 可以用于数据序列化，但您必须制定一个协议 您拥有您在客户端和服务器之间交换的消息，这 您可以提前知道要读取的数据量以进行 unpickling 的方式（请参阅 下面）
• 对于recv，如果您收到大块，您将获得更好的性能，因此将 80 替换为 4096 甚至更多
• 注意sys.getsizeof：它返回内存中对象的大小，不是 与通过网络发送的字节的大小（长度）相同；为一个 Python字符串这两个值根本不一样
• 请注意您发送的帧的大小。下面的代码支持最大为 65535 的帧。如果您有更大的帧，请将“H”更改为“L”。

协议示例：

client_cv.py

import cv2
import numpy as np
import socket
import sys
import pickle
import struct ### new code
cap=cv2.VideoCapture(0)
clientsocket=socket.socket(socket.AF_INET,socket.SOCK_STREAM)
clientsocket.connect(('localhost',8089))
while True:
    ret,frame=cap.read()
    data = pickle.dumps(frame) ### new code
    clientsocket.sendall(struct.pack("H", len(data))+data) ### new code

server_cv.py

import socket
import sys
import cv2
import pickle
import numpy as np
import struct ## new

HOST=''
PORT=8089

s=socket.socket(socket.AF_INET,socket.SOCK_STREAM)
print 'Socket created'

s.bind((HOST,PORT))
print 'Socket bind complete'
s.listen(10)
print 'Socket now listening'

conn,addr=s.accept()

### new
data = ""
payload_size = struct.calcsize("H") 
while True:
    while len(data) < payload_size:
        data += conn.recv(4096)
    packed_msg_size = data[:payload_size]
    data = data[payload_size:]
    msg_size = struct.unpack("H", packed_msg_size)[0]
    while len(data) < msg_size:
        data += conn.recv(4096)
    frame_data = data[:msg_size]
    data = data[msg_size:]
    ###

    frame=pickle.loads(frame_data)
    print frame
    cv2.imshow('frame',frame)

您可能可以对所有这些进行很多优化（减少复制，使用缓冲区接口等），但至少您可以理解。

Answer 2

经过几个月的互联网搜索，这是我想出的，我已经将它整齐地打包成类，单元测试和文档为SmoothStream 看看，这是我唯一简单且有效的流媒体版本到处都能找到。

我使用了这段代码并将我的代码包裹起来。

查看器.py

import cv2
import zmq
import base64
import numpy as np

context = zmq.Context()
footage_socket = context.socket(zmq.SUB)
footage_socket.bind('tcp://*:5555')
footage_socket.setsockopt_string(zmq.SUBSCRIBE, np.unicode(''))

while True:
    try:
        frame = footage_socket.recv_string()
        img = base64.b64decode(frame)
        npimg = np.fromstring(img, dtype=np.uint8)
        source = cv2.imdecode(npimg, 1)
        cv2.imshow("Stream", source)
        cv2.waitKey(1)

    except KeyboardInterrupt:
        cv2.destroyAllWindows()
        break

Streamer.py

import base64
import cv2
import zmq

context = zmq.Context()
footage_socket = context.socket(zmq.PUB)
footage_socket.connect('tcp://localhost:5555')

camera = cv2.VideoCapture(0)  # init the camera

while True:
    try:
        grabbed, frame = camera.read()  # grab the current frame
        frame = cv2.resize(frame, (640, 480))  # resize the frame
        encoded, buffer = cv2.imencode('.jpg', frame)
        jpg_as_text = base64.b64encode(buffer)
        footage_socket.send(jpg_as_text)

    except KeyboardInterrupt:
        camera.release()
        cv2.destroyAllWindows()
        break

Answer 3

我将代码从 @mguijarr 更改为使用 Python 3。对代码所做的更改：

• data 现在是 byte literal 而不是字符串文字
• 将“H”更改为“L”以发送更大的帧大小。基于the documentation，我们现在可以发送大小为 2^32 的帧，而不仅仅是 2^16。

服务器.py

import pickle
import socket
import struct

import cv2

HOST = ''
PORT = 8089

s = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
print('Socket created')

s.bind((HOST, PORT))
print('Socket bind complete')
s.listen(10)
print('Socket now listening')

conn, addr = s.accept()

data = b'' ### CHANGED
payload_size = struct.calcsize("L") ### CHANGED

while True:

    # Retrieve message size
    while len(data) < payload_size:
        data += conn.recv(4096)

    packed_msg_size = data[:payload_size]
    data = data[payload_size:]
    msg_size = struct.unpack("L", packed_msg_size)[0] ### CHANGED

    # Retrieve all data based on message size
    while len(data) < msg_size:
        data += conn.recv(4096)

    frame_data = data[:msg_size]
    data = data[msg_size:]

    # Extract frame
    frame = pickle.loads(frame_data)

    # Display
    cv2.imshow('frame', frame)
    cv2.waitKey(1)

客户端.py

import cv2
import numpy as np
import socket
import sys
import pickle
import struct

cap=cv2.VideoCapture(0)
clientsocket=socket.socket(socket.AF_INET,socket.SOCK_STREAM)
clientsocket.connect(('localhost',8089))

while True:
    ret,frame=cap.read()
    # Serialize frame
    data = pickle.dumps(frame)

    # Send message length first
    message_size = struct.pack("L", len(data)) ### CHANGED

    # Then data
    clientsocket.sendall(message_size + data)

Answer 4

我有点晚了，但我强大的线程VidGear 视频处理python 库现在提供NetGear API，它专门用于通过网络在互连系统之间实时同步传输视频帧。这是一个例子：

A.服务器端：（Bare-Minimum 示例）

打开您喜欢的终端并执行以下 python 代码：

注意： 您可以随时在服务器端和客户端通过在服务器端按键盘上的 [Ctrl+c] 来结束流式传输！

# import libraries
from vidgear.gears import VideoGear
from vidgear.gears import NetGear

stream = VideoGear(source='test.mp4').start() #Open any video stream
server = NetGear() #Define netgear server with default settings

# infinite loop until [Ctrl+C] is pressed
while True:
    try: 
        frame = stream.read()
        # read frames

        # check if frame is None
        if frame is None:
            #if True break the infinite loop
            break

        # do something with frame here

        # send frame to server
        server.send(frame)

    except KeyboardInterrupt:
        #break the infinite loop
        break

# safely close video stream
stream.stop()
# safely close server
writer.close()

B.客户端：（最简单的例子）

然后在同一系统上打开另一个终端并执行以下python代码并查看输出：

# import libraries
from vidgear.gears import NetGear
import cv2

#define netgear client with `receive_mode = True` and default settings
client = NetGear(receive_mode = True)

# infinite loop
while True:
    # receive frames from network
    frame = client.recv()

    # check if frame is None
    if frame is None:
        #if True break the infinite loop
        break

    # do something with frame here

    # Show output window
    cv2.imshow("Output Frame", frame)

    key = cv2.waitKey(1) & 0xFF
    # check for 'q' key-press
    if key == ord("q"):
        #if 'q' key-pressed break out
        break

# close output window
cv2.destroyAllWindows()
# safely close client
client.close()

更高级的用法和相关文档可以在这里找到：https://github.com/abhiTronix/vidgear/wiki/NetGear

Answer 5

正如@Rohan Sawant 所说，我使用 zmq 库而不使用 base64 编码。这是新代码

Streamer.py

import base64
import cv2
import zmq
import numpy as np
import time

context = zmq.Context()
footage_socket = context.socket(zmq.PUB)
footage_socket.connect('tcp://192.168.1.3:5555')

camera = cv2.VideoCapture(0)  # init the camera

while True:
        try:
                grabbed, frame = camera.read()  # grab the current frame
                frame = cv2.resize(frame, (640, 480))  # resize the frame
                encoded, buffer = cv2.imencode('.jpg', frame)
                footage_socket.send(buffer)


        except KeyboardInterrupt:
                camera.release()
                cv2.destroyAllWindows()
                break

查看器.py

import cv2
import zmq
import base64
import numpy as np

context = zmq.Context()
footage_socket = context.socket(zmq.SUB)
footage_socket.bind('tcp://*:5555')
footage_socket.setsockopt_string(zmq.SUBSCRIBE, np.unicode(''))

while True:
    try:
        frame = footage_socket.recv()
        npimg = np.frombuffer(frame, dtype=np.uint8)
        #npimg = npimg.reshape(480,640,3)
        source = cv2.imdecode(npimg, 1)
        cv2.imshow("Stream", source)
        cv2.waitKey(1)

    except KeyboardInterrupt:
        cv2.destroyAllWindows()
        break

Answer 6

最近我发布了 imagiz 包，用于使用 OpenCV 和 ZMQ 通过网络进行快速且无阻塞的实时视频流。

https://pypi.org/project/imagiz/

客户：

import imagiz
import cv2


client=imagiz.Client("cc1",server_ip="localhost")
vid=cv2.VideoCapture(0)
encode_param = [int(cv2.IMWRITE_JPEG_QUALITY), 90]

while True:
    r,frame=vid.read()
    if r:
        r, image = cv2.imencode('.jpg', frame, encode_param)
        client.send(image)
    else:
        break

服务器：

import imagiz
import cv2

server=imagiz.Server()
while True:
    message=server.recive()
    frame=cv2.imdecode(message.image,1)
    cv2.imshow("",frame)
    cv2.waitKey(1)

Answer 7

我已经让它在我的 MacOS 上运行。

我使用来自@mguijarr 的代码并将 struct.pack 从“H”更改为“L”。

# Server.py:
import socket
import sys
import cv2
import pickle
import numpy as np
import struct ## new


HOST=''
PORT=8089
    
s=socket.socket(socket.AF_INET,socket.SOCK_STREAM)
print 'Socket created'
    
s.bind((HOST,PORT))
print 'Socket bind complete'
s.listen(10)
print 'Socket now listening'
    
conn,addr=s.accept()
    
# new
data = ""
payload_size = struct.calcsize("L") 
while True:
    while len(data) < payload_size:
        data += conn.recv(4096)
    packed_msg_size = data[:payload_size]
    data = data[payload_size:]
    msg_size = struct.unpack("L", packed_msg_size)[0]
    while len(data) < msg_size:
        data += conn.recv(4096)
    frame_data = data[:msg_size]
    data = data[msg_size:]
    
    
    frame=pickle.loads(frame_data)
    print frame
    cv2.imshow('frame',frame)
        
    key = cv2.waitKey(10)
    if (key == 27) or (key == 113):
        break
    
cv2.destroyAllWindows()

# Client.py
import cv2
import numpy as np
import socket
import sys
import pickle
import struct ### new code


cap=cv2.VideoCapture(0)
clientsocket=socket.socket(socket.AF_INET,socket.SOCK_STREAM)
clientsocket.connect(('localhost',8089))

while True:
    ret,frame=cap.read()
    data = pickle.dumps(frame) ### new code
    clientsocket.sendall(struct.pack("L", len(data))+data) ### new code