Introduction

Detecting the real-time emotion of the person with a camera input is one of the advanced features in the machine learning process. The detection of emotion of a person using a camera is useful for various research and analytics purposes. The detection of emotion is made by using the machine learning concept. You can use the trained dataset to detect the emotion of the human being. For detecting the different emotions, first, you need to train those different emotions, or you can use a dataset already available on the internet. In this article, we will discuss creating a Python program to detect the real-time emotion of a human being using the camera.

Installing Dependencies

For using this machine learning concept, you need to install a lot of dependencies into your system using the command prompt. The machine learning algorithm used by me was a tensor flow algorithm, which was designed by Google for machine learning functions. For analyzing faces. you need to detect the faces, to know more about detecting faces using python, you can refer to my article by clicking here. You need a cascade file for this process, you can download it from my git-hub page or in the download section.
You can install the dependencies by using the commands given below:
  1. pip install opencv-python

  3. pip install tensorflow
  5. pip install numpy
  7. pip install pandas
  9. pip install keras
  11. pip install adam
  13. pip install kwargs
  15. pip install cinit

Training the Dataset

For training purposes, I use the predefined un trained dataset CSV file as my main input for my input for training the machine. You can use the code given below for training the machine using the dataset. Before that, you need to ensure that all required files in the same repository where the program presents otherwise it will through some error. You can download the data set by clicking here.
  1. import sys, os
  2. import pandas as pd
  3. import numpy as np
  5. from keras.models import Sequential
  6. from keras.layers import Dense, Dropout, Activation, Flatten
  7. from keras.layers import Conv2D, MaxPooling2D, BatchNormalization,AveragePooling2D
  8. from keras.losses import categorical_crossentropy
  9. from keras.optimizers import Adam
  10. from keras.regularizers import l2
  11. from keras.utils import np_utils
  12. # pd.set_option('display.max_rows', 500)
  13. # pd.set_option('display.max_columns', 500)
  14. # pd.set_option('display.width', 1000)
  16. df=pd.read_csv('fer2013.csv')
  18. # print(df.info())
  19. # print(df["Usage"].value_counts())
  21. # print(df.head())
  22. X_train,train_y,X_test,test_y=[],[],[],[]
  24. for index, row in df.iterrows():
  25. val=row['pixels'].split(" ")
  26. try:
  27. if 'Training' in row['Usage']:
  28. X_train.append(np.array(val,'float32'))
  29. train_y.append(row['emotion'])
  30. elif 'PublicTest' in row['Usage']:
  31. X_test.append(np.array(val,'float32'))
  32. test_y.append(row['emotion'])
  33. except:
  34. print(f"error occured at index :{index} and row:{row}")
  37. num_features = 64
  38. num_labels = 7
  39. batch_size = 64
  40. epochs = 30
  41. width, height = 48, 48
  44. X_train = np.array(X_train,'float32')
  45. train_y = np.array(train_y,'float32')
  46. X_test = np.array(X_test,'float32')
  47. test_y = np.array(test_y,'float32')
  49. train_y=np_utils.to_categorical(train_y, num_classes=num_labels)
  50. test_y=np_utils.to_categorical(test_y, num_classes=num_labels)
  52. #cannot produce
  53. #normalizing data between oand 1
  54. X_train -= np.mean(X_train, axis=0)
  55. X_train /= np.std(X_train, axis=0)
  57. X_test -= np.mean(X_test, axis=0)
  58. X_test /= np.std(X_test, axis=0)
  60. X_train = X_train.reshape(X_train.shape[0], 48, 48, 1)
  62. X_test = X_test.reshape(X_test.shape[0], 48, 48, 1)
  64. # print(f"shape:{X_train.shape}")
  65. ##designing the cnn
  66. #1st convolution layer
  67. model = Sequential()
  69. model.add(Conv2D(64, kernel_size=(3, 3), activation='relu', input_shape=(X_train.shape[1:])))
  70. model.add(Conv2D(64,kernel_size= (3, 3), activation='relu'))
  71. # model.add(BatchNormalization())
  72. model.add(MaxPooling2D(pool_size=(2,2), strides=(2, 2)))
  73. model.add(Dropout(0.5))
  75. #2nd convolution layer
  76. model.add(Conv2D(64, (3, 3), activation='relu'))
  77. model.add(Conv2D(64, (3, 3), activation='relu'))
  78. # model.add(BatchNormalization())
  79. model.add(MaxPooling2D(pool_size=(2,2), strides=(2, 2)))
  80. model.add(Dropout(0.5))
  82. #3rd convolution layer
  83. model.add(Conv2D(128, (3, 3), activation='relu'))
  84. model.add(Conv2D(128, (3, 3), activation='relu'))
  85. # model.add(BatchNormalization())
  86. model.add(MaxPooling2D(pool_size=(2,2), strides=(2, 2)))
  88. model.add(Flatten())
  90. #fully connected neural networks
  91. model.add(Dense(1024, activation='relu'))
  92. model.add(Dropout(0.2))
  93. model.add(Dense(1024, activation='relu'))
  94. model.add(Dropout(0.2))
  96. model.add(Dense(num_labels, activation='softmax'))
  98. # model.summary()
  100. #Compliling the model
  101. model.compile(loss=categorical_crossentropy,
  102. optimizer=Adam(),
  103. metrics=['accuracy'])
  105. #Training the model
  106. model.fit(X_train, train_y,
  107. batch_size=batch_size,
  108. epochs=epochs,
  109. verbose=1,
  110. validation_data=(X_test, test_y),
  111. shuffle=True)
  114. #Saving the model to use it later on
  115. fer_json = model.to_json()
  116. with open("fer.json", "w") as json_file:
  117. json_file.write(fer_json)
  118. model.save_weights("fer.h5")

Detecting Real-Time Emotion

For detecting the emotion, first, you need to run the train.py program to train the data. Then you can use the code given below:
  1. import os
  2. import cv2
  3. import numpy as np
  4. from keras.models import model_from_json
  5. from keras.preprocessing import image
  7. #load model
  8. model = model_from_json(open("fer.json", "r").read())
  9. #load weights
  10. model.load_weights('fer.h5')
  13. face_haar_cascade = cv2.CascadeClassifier('haarcascade_frontalface_default.xml')
  16. cap=cv2.VideoCapture(0)
  18. while True:
  19. ret,test_img=cap.read()# captures frame and returns boolean value and captured image
  20. if not ret:
  21. continue
  22. gray_img= cv2.cvtColor(test_img, cv2.COLOR_BGR2GRAY)
  24. faces_detected = face_haar_cascade.detectMultiScale(gray_img, 1.32, 5)
  27. for (x,y,w,h) in faces_detected:
  28. cv2.rectangle(test_img,(x,y),(x+w,y+h),(255,0,0),thickness=7)
  29. roi_gray=gray_img[y:y+w,x:x+h]#cropping region of interest i.e. face area from image
  30. roi_gray=cv2.resize(roi_gray,(48,48))
  31. img_pixels = image.img_to_array(roi_gray)
  32. img_pixels = np.expand_dims(img_pixels, axis = 0)
  33. img_pixels /= 255
  35. predictions = model.predict(img_pixels)
  37. #find max indexed array
  38. max_index = np.argmax(predictions[0])
  40. emotions = ('angry', 'disgust', 'fear', 'happy', 'sad', 'surprise', 'neutral')
  41. predicted_emotion = emotions[max_index]
  43. cv2.putText(test_img, predicted_emotion, (int(x), int(y)), cv2.FONT_HERSHEY_SIMPLEX, 1, (0,0,255), 2)
  45. resized_img = cv2.resize(test_img, (1000, 700))
  46. cv2.imshow('Facial emotion analysis ',resized_img)
  50. if cv2.waitKey(10) == ord('q'):#wait until 'q' key is pressed
  51. break
  53. cap.release()
  54. cv2.destroyAllWindows

Output Verification

Now you can run the videoTester.py program. Your camera automatically turns on and detects the emotion of your face.

Conclusion

This is just a beginning step in face detection. You can download the program files from my git-hub link by clicking here. Feel free to use this for future enhancements.