Question

我目前有两个脚本，一个脚本捕获素材并发布numpy数组值（servant.py），另一个脚本（master.py）然后使用opencv处理这些值以用于以后的面部识别实现。问题在于，由于发送的Internet包到达主脚本的时间非常延迟，因此现在它非常慢。我想知道是否有更好的方法？当然，现在每秒仅发送一帧，因此我需要脚本能够每秒处理24帧。

这是两个脚本：

master.py

import paho.mqtt.client as mqtt
import numpy as np
import json
import PIL

MQTT_SERVER = "iot.eclipse.org"
MQTT_PATH = "test_channel"

def on_connect(client, userdata, flags, rc):
    print("connected with result code " + str(rc))
    client.subscribe(MQTT_PATH)

def on_message(client, userdata, msg):
    dtype = "uint8"
    dshape = (480, 640, 3)
    data = msg.payload
    img_array = np.fromstring(data, dtype=dtype).reshape(dshape)
    img = PIL.Image.fromarray(img_array)
    img.save("img.png")
    img.show()


client = mqtt.Client()
client.on_connect = on_connect
client.on_message = on_message
client.connect(MQTT_SERVER, 1883, 60)

client.loop_forever()

servant.py

import paho.mqtt.client as mqtt
import time
import cv2
import numpy
import json

MQTT_SERVER = "iot.eclipse.org"
MQTT_PATH = "test_channel"

mqttc = mqtt.Client()
mqttc.connect(MQTT_SERVER, 1883, 60)

cap = cv2.VideoCapture(0)

while True:
    ret, frame = cap.read()
    MQTT_MESSAGE = frame.tostring()
    mqttc.publish(MQTT_PATH, MQTT_MESSAGE)
    mqttc.loop()
    time.sleep(1)

Answer 1

我认为您最好的选择是改用cvlc或gstreamer之类的东西从相机中流式传输，然后从opencv中简单地连接到它（cap = cv2.VideoCapture（“ http://some_url:some_port”）），然后您应该始终根据带宽，编解码器等获得合理的吞吐量。

Answer 2

使用WebcamVideoStream中的imutils，将图像压缩为jpg，并将其发布为base64：

import base64
import cv2
from imutils.video import WebcamVideoStream

cap = WebcamVideoStream(0)
cap.stream.set(cv2.CAP_PROP_FRAME_WIDTH, 640)
cap.stream.set(cv2.CAP_PROP_FRAME_HEIGHT, 480)
cap.start()

while True:
    frame = cap.read()
    if frame is None:
        print("Continue")
        continue

    retval, buffer = cv2.imencode('.jpg', frame)
    jpg_as_text = base64.b64encode(buffer).decode('utf-8')
    publish(jpg_as_text)

Python MQTT提高了图像numpy数组的发布速度

2 个答案: