test.py

import numpy as np
import cv2
import tensorflow as tf

face_detection = cv2.CascadeClassifier('haar_cascade_face_detection.xml')

camera = cv2.VideoCapture(0)
camera.set(cv2.CAP_PROP_FRAME_WIDTH, 1024)
camera.set(cv2.CAP_PROP_FRAME_HEIGHT, 768)
settings = {
    'scaleFactor': 1.3, 
    'minNeighbors': 5, 
    'minSize': (50, 50)
}

labels = ["Neutral","Happy","Sad","Surprise","Angry"]

model = tf.keras.models.load_model('expression.model')

while True:
	ret, img = camera.read()
	gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
	detected = face_detection.detectMultiScale(gray, **settings)
    
	for x, y, w, h in detected:
		cv2.rectangle(img, (x, y), (x+w, y+h), (245, 135, 66), 2)
		cv2.rectangle(img, (x, y), (x+w//3, y+20), (245, 135, 66), -1)
		face = gray[y+5:y+h-5, x+20:x+w-20]
		face = cv2.resize(face, (48,48)) 
		face = face/255.0
		
		predictions = model.predict(np.array([face.reshape((48,48,1))])).argmax()
		state = labels[predictions]
		font = cv2.FONT_HERSHEY_SIMPLEX
		cv2.putText(img,state,(x+10,y+15), font, 0.5, (255,255,255), 2, cv2.LINE_AA)
		
	cv2.imshow('Facial Expression', img)
	img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
	# writer.writeFrame(img)
	if cv2.waitKey(5) != -1:
		break

camera.release()
cv2.destroyAllWindows()