如何让树莓派OpenCV人脸检测仅发送一次MQTT消息？

Hey there! I get it—your current script sends an MQTT message every single frame a face is detected, which leads to a flood of unwanted messages. Let's fix this with a simple state-tracking variable that ensures we only send one notification per face appearance, and reset the ability to send once the face is gone.

The Core Idea #

We'll add a boolean variable (let's call it face_already_notified) that keeps track of whether we've already sent the MQTT message for the current detected face. Here's how it works:

• Initialize the variable as False (we haven't notified anyone yet).
• When a face is detected:
  • If face_already_notified is False, send the MQTT message and set the variable to True.
  • If it's already True, do nothing (we've already notified for this face).
• When no face is detected, reset face_already_notified back to False so we can send a message again the next time a face appears.

Modified Code #

Here's your updated script with the fix included (I've added comments to highlight key changes):

from picamera.array import PiRGBArray
from picamera import PiCamera
import paho.mqtt.client as mqtt
import time
import cv2

# MQTT setup
host = '10.0.0.192'
client = mqtt.Client("Security")
client.connect(host)

# Image processing parameters
threshold = 60 # BINARY threshold
blurValue = 41 # GaussianBlur parameter
bgSubThreshold = 50

# NEW: State variable to track if we've already sent the notification
face_already_notified = False

# Camera initialization
camera = PiCamera()
camera.resolution = (640, 480)
camera.framerate = 64
rawCapture = PiRGBArray(camera, size=(640, 480))

# Allow camera to warmup
time.sleep(0.1)

# IMPORTANT: Move cascade classifier initialization outside the loop!
# Loading it every frame wastes resources and slows down your script
face_cascade = cv2.CascadeClassifier('/home/pi/opencv-face-sentdex/faces.xml')

# Capture frames from the camera
for frame in camera.capture_continuous(rawCapture, format="bgr", use_video_port=True):
    image = frame.array
    gray = cv2.cvtColor(image,cv2.COLOR_BGR2GRAY)
    blur = cv2.GaussianBlur(gray, (blurValue, blurValue), 0)
    faces = face_cascade.detectMultiScale(blur, 1.1, 5)

    # Check if faces are detected
    if len(faces) > 0:
        # Draw rectangle around each face
        for (x,y,w,h) in faces:
            cv2.rectangle(image,(x,y),(x+w,y+h),(255,255,0),2)
        
        # NEW: Only send MQTT message if we haven't already for this face
        if not face_already_notified:
            client.publish("/home/pi/snips-app","faceDetected")#publish
            face_already_notified = True
            print("Face detected - sent MQTT message") # Optional: for debugging
    else:
        # NEW: Reset the notification state when no face is detected
        if face_already_notified:
            face_already_notified = False
            print("Face lost - reset notification state") # Optional: for debugging

    # Show the frame
    cv2.imshow("Frame", image)
    key = cv2.waitKey(1) & 0xFF

    # Clear the stream for next frame
    rawCapture.truncate(0)

    # Exit on 'q' press
    if key == ord("q"):
        break

# Cleanup resources
cv2.destroyAllWindows()
client.disconnect()

Key Improvements Beyond the Fix #

I also added one critical optimization:

• Moved the face_cascade = cv2.CascadeClassifier(...) line outside the loop. Loading the classifier every frame is unnecessary and wastes processing power—loading it once at the start makes your script run much smoother.

How It Works #

• When a face first appears, face_already_notified is False, so we send the message and flip the variable to True.
• As long as the face stays in the frame, face_already_notified remains True, so no duplicate messages are sent.
• When the face leaves the frame, we reset face_already_notified to False, so the next time a face is detected, we'll send a fresh notification.

This way, you get exactly one MQTT alert per face appearance, no matter how long the face stays in view.

内容的提问来源于stack exchange，提问作者Nick D