Dlib を用いて，顔検出，顔のランドマーク検知（68 ランドマーク法），表情判定を行う（Windows 上）

import numpy as np
import cv2
from tensorflow.keras.preprocessing import image
import dlib
from imutils import face_utils
import imutils
from sklearn import preprocessing
import math
from keras.models import model_from_json
#-----------------------------
#opencv initialization
face_cascade = cv2.CascadeClassifier('haarcascade_frontalface_default.xml')
cap = cv2.VideoCapture(0)

#-----------------------------
#face expression recognizer initialization
# Using pretrained model
model = model_from_json(open("model/model.json", "r").read())
model.load_weights('model/model.h5') #load weights

#-----------------------------

emotions = ( 'Angry' , 'Disgust' , 'Fear' , 'Happy'  , 'Neutral' ,  'Sad' , 'Surprise')
# initialize dlib's face detector and create a predictor
detector = dlib.get_frontal_face_detector()
predictor = dlib.shape_predictor("shape_predictor_68_face_landmarks.dat")


def detect_parts(image):
    distances = []
    # resize the image, and convert it to grayscale
    image = imutils.resize(image, width=200, height=200)
    
    gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
    # detect faces in the grayscale image
    rects = detector(gray, 1)
    
    # loop over the face detections
    for (i, rect) in enumerate(rects):
        shape = predictor(gray, rect)
        shape = face_utils.shape_to_np(shape)
        distances = euclidean_all(shape)
        # visualize all facial landmarks with a transparent overlay
        #output = face_utils.visualize_facial_landmarks(image, shape)
        #cv2.imshow("Image", output)
        #cv2.waitKey(0)    
    return distances

def euclidean(a, b):
    dist = math.sqrt(math.pow((b[0] - a[0]), 2) + math.pow((b[1] - a[1]), 2))
    return dist 

# calculates distances between all 68 elements
def euclidean_all(a):  
    distances = ""
    for i in range(0, len(a)):
        for j in range(0, len(a)):
            dist = euclidean(a[i], a[j])
            dist = "%.2f" % dist;
            distances = distances + " " + str(dist)
    return distances


def box_label(bgr, x1, y1, x2, y2, label): 
    cv2.rectangle(bgr, (x1, y1), (x2, y2), (255, 0, 0), 1, 1)
    cv2.rectangle(bgr, (int(x1), int(y1-25)), (x2, y1), (255,255,255), -1)
    cv2.putText(bgr, label, (x1, int(y1-5)), cv2.FONT_HERSHEY_COMPLEX, 0.7, (0,0,0), 1)


while(True):
    ret, img = cap.read()
    gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
    faces = face_cascade.detectMultiScale(gray, 1.3, 5)
    for (x,y,w,h) in faces:
        detected_face = img[int(y):int(y+h), int(x):int(x+w)] #crop detected face
        distances = detect_parts(detected_face)
        if(len(distances)!=0):
            val = distances.split(" ")[1:]
            val = np.array(val)
            val = val.astype(np.float)
            val = np.expand_dims(val, axis = 1)            
            minmax = preprocessing.MinMaxScaler()
            val = minmax.fit_transform(val)
            val = val.reshape(1,4624)
            predictions = model.predict(val) #store probabilities of 6 expressions
        #find max indexed array ( 'Angry' , 'Disgust' , 'Fear' , 'Happy'  , 'Neutral' ,  'Sad' , 'Surprise')
            print ("Angry: %", predictions[0][0]/1.0 * 100)
            print ("Disgust: %", predictions[0][1]/1.0 * 100)
            print ("Fear: %", predictions[0][2]/1.0 * 100)
            print ("Happy: %", predictions[0][3]/1.0 * 100)
            print ("Neutral: %", predictions[0][4]/1.0 * 100)
            print ("Sad: %", predictions[0][5]/1.0 * 100)    
            print ("Surprised: %", predictions[0][6]/1.0 * 100)        
            print ("----------------------"    )    
            max_index = np.argmax(predictions[0])
            emotion = emotions[max_index]
            #write emotion text above rectangle
        box_label(img, x, y, x+w, y+h, emotion+":"+'{:2.2f}'.format(np.max(predictions[0])/1.0 * 100))
    cv2.imshow('img',img)
    if cv2.waitKey(1) & 0xFF == ord('q'): #press q to quit
        break

#kill open cv things        
cap.release()
cv2.destroyAllWindows()

※ 途中で止めたいとき，右上の「x」をクリックしない．画面の中をクリックしてから，「q」のキーを押して閉じる

Dlib を用いて，顔検出，顔のランドマーク検知（68 ランドマーク法），表情判定を行う（Windows 上）

前準備

Build Tools for Visual Studio 2022 （ビルドツール for Visual Studio 2022）のインストール

Git のインストール

7-Zip のインストール

Python 64 ビット版のインストール，pip と setuptools の更新

Dlib のインストール

Dlib のソースコード等と，Dlib の学習済みモデルのダウンロード

ezgiakcora/Facial-Expression-Keras を動かしてみる