얼굴 인식 장치

** TensorFlow 2.0부터는 라이브러리 구조가 변경되어 아래 코드를 바로 사용할 수 없음.

 

1. AI_PI.py

 - 사전에 학습된 가중치 파일과 모델을 라즈베리파이에 설치하여 작동

 - 판단 이후는 GPIO를 통하여 외부 장치(LED) 등으로 결과를 알려줌 (미구현)

# python 3.4
from socket import *
from matplotlib.image import imread
import tensorflow as tf
import numpy as np

# SOCKET
HOST = ''
PORT = 12345
SIZE = 2048
MSGLEN = 0

# NETWORK
learning_rate = 0.001
n_input = 30000
n_classes = 2
dropout = 0.75

x = tf.placeholder(tf.float32, [None, n_input])
y = tf.placeholder(tf.float32, [None, n_classes])
keep_prob = tf.placeholder(tf.float32)

weights ={
    'wc1' : tf.get_variable("wc1", shape=[3, 3, 3, 8], initializer =tf.contrib.layers.xavier_initializer()),
    'wc2' : tf.get_variable("wc2", shape=[3, 3, 8, 16], initializer =tf.contrib.layers.xavier_initializer()),
    'wc3' : tf.get_variable("wc3", shape=[3, 3, 16, 32], initializer =tf.contrib.layers.xavier_initializer()),
    'wc4' : tf.get_variable("wc4", shape=[3, 3, 32, 64], initializer =tf.contrib.layers.xavier_initializer()),
    'wc5' : tf.get_variable("wc5", shape=[3, 3, 64, 128], initializer =tf.contrib.layers.xavier_initializer()),
    'wd1' : tf.get_variable("wd1", shape=[1*1*128, 1024], initializer =tf.contrib.layers.xavier_initializer()),
    'wout' : tf.get_variable("wout", shape=[1024, n_classes], initializer =tf.contrib.layers.xavier_initializer())
    }

biases = {
    'bc1' : tf.Variable(tf.zeros([8])),
    'bc2' : tf.Variable(tf.zeros([16])),
    'bc3' : tf.Variable(tf.zeros([32])),
    'bc4' : tf.Variable(tf.zeros([64])),
    'bc5' : tf.Variable(tf.zeros([128])),
    'bd1' : tf.Variable(tf.zeros([1024])),
    'bout': tf.Variable(tf.zeros([n_classes]))
    }


def conv2d(img, w, b):
    return tf.nn.relu(tf.nn.bias_add(tf.nn.conv2d(img, w, strides=[1, 1, 1, 1], padding='VALID'), b))

def max_pool(img, k):
    return tf.nn.max_pool(img, ksize=[1, k, k, 1], strides=[1, k, k, 1], padding='VALID')


def conv_net(x, weights, biases, dropout):
    _X = tf.reshape(x, shape=[-1, 100, 100, 3])
    
    conv1 = conv2d(_X, weights['wc1'], biases['bc1'])
    conv1 = max_pool(conv1, k=2)
    conv1 = tf.nn.dropout(conv1, dropout)

    conv2 = conv2d(conv1, weights['wc2'], biases['bc2'])
    conv2 = max_pool(conv2, k=2)
    conv2 = tf.nn.dropout(conv2, dropout)

    conv3 = conv2d(conv2, weights['wc3'], biases['bc3'])
    conv3 = max_pool(conv3, k=2)
    conv3 = tf.nn.dropout(conv3, dropout)

    conv4 = conv2d(conv3, weights['wc4'], biases['bc4'])
    conv4 = max_pool(conv4, k=2)
    conv4 = tf.nn.dropout(conv4, dropout)

    conv5 = conv2d(conv4, weights['wc5'], biases['bc5'])
    conv5 = max_pool(conv5, k=2)
    conv5 = tf.nn.dropout(conv5, dropout)

    dense1 = tf.reshape(conv5, [-1, weights['wd1'].get_shape().as_list()[0]])
    dense1 = tf.nn.relu(tf.add(tf.matmul(dense1, weights['wd1']), biases['bd1']))
    dense1 = tf.nn.dropout(dense1, dropout)

    out = tf.add(tf.matmul(dense1, weights['wout']), biases['bout'])

    return out

def autoConnection():
    s_broad = socket(AF_INET, SOCK_DGRAM)
    s_broad.connect(('8.8.8.8', 0))
    priip = s_broad.getsockname()[0]
    s_broad.setsockopt(SOL_SOCKET, SO_REUSEADDR, 1)
    s_broad.setsockopt(SOL_SOCKET, SO_BROADCAST, 1)
    
    s_broad.sendto(priip.encode(), ('255.255.255.255', 8000))
    print('ip : ', priip)
    s_broad.close()

def testNetwork():
    print('Start server')
    s = socket(AF_INET, SOCK_STREAM)
    s.setsockopt(SOL_SOCKET, SO_REUSEADDR, 1)
    s.bind((HOST, PORT))
    s.listen(1)
    
    autoConnection()

    s, addr = s.accept()
    print('conneted ... ' , addr)

    # File transfer
    print('File receiving')
    filesize = int(s.recv(4).decode())
    total = 0
    
    f = open('recv.jpg', 'wb')
    while 1:
        data = s.recv(2048)
        if data:
            total += len(data)
            f.write(data)
            if(total == filesize):
                break
    f.close()
    
    # Session run
    print('Start Network')
    img = imread('recv.jpg')
    float_img = img.astype(np.float32)
    float_img *= 1.0/255.0
    float_img = float_img.reshape([1, 100 * 100 * 3])

    pred = conv_net(x, weights, biases, keep_prob)
    cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=pred, labels=y))
    optimizer = tf.train.AdamOptimizer(learning_rate=learning_rate).minimize(cost)
    
    correct_pred = tf.equal(tf.argmax(pred, 1), tf.argmax(y, 1))
    accuracy = tf.reduce_mean(tf.cast(correct_pred, tf.float32))

    init = tf.global_variables_initializer()
    saver = tf.train.Saver()
    
    with tf.Session() as sess:
        sess.run(init)
        saver.restore(sess, 'tmp/model.ckpt')

        sess.run(optimizer, feed_dict=)
        prd, acc = sess.run([pred, accuracy], feed_dict=)
        
        if acc > 0.9:
            s.send(b'1') # hwan
        else:
            s.send(b'0') # other

    s.close()

testNetwork()

print('exit')
exit()

 

2. AI_PC.py

 - 라즈베리 파이에서 PyCamera가 OpenCV로 인식되지 않아 노트북카메라와 소켓을 활용함

   (차후 수정 시 없어지는 부분)

 - 카메라에 사람 얼굴이 인식되면 캡쳐 후 전송 (Haar Cascade 인식기를 사용)

 - 동일 네트워크 상에 라즈베리파이가 존재하면 자동 연결

# python3.5
import cv2
import numpy as np
from socket import *
import os
import time

# Haar cascade load
face_cascade = cv2.CascadeClassifier('haarcascade_frontalface_default.xml')
eye_cascade = cv2.CascadeClassifier('haarcascade_eye.xml')
IMAGE_NAME = 'face.jpg'

# face detection & capture
def face_detect():
    img = cv2.VideoCapture(0)
    while img.isOpened():
        ret, frame = img.read()

        gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
        faces = face_cascade.detectMultiScale(gray, 1.3, 5)
        if len(faces) > 0:
            print('\nFace Detaction! -> ', end='')
        
            for (x, y, w, h) in faces:
                crop = frame[y:y+h, x:x+h] 
                roi_gray = gray[y:y+h, x:x+h]
        
                eyes = eye_cascade.detectMultiScale(roi_gray)
                if len(eyes) > 0:
                    crop[:, :, 0] = cv2.equalizeHist(crop[:, :, 0])
                    crop[:, :, 1] = cv2.equalizeHist(crop[:, :, 1])
                    crop[:, :, 2] = cv2.equalizeHist(crop[:, :, 2])
                    out_img = cv2.resize(crop, (100, 100), interpolation = cv2.INTER_CUBIC)
                    cv2.imwrite(IMAGE_NAME, out_img)
                    print('Capture!\n')
                    return 1
    return 0

# NETWORK
print('Start Client')
s_broad = socket(AF_INET, SOCK_DGRAM)
s_broad.bind(('', 8000))
msg, addr = s_broad.recvfrom(1024)
HOST = msg.decode()
PORT = 12345
print('Server IP :', HOST)

s = socket(AF_INET, SOCK_STREAM)
print('Server connect ...', end='')
time.sleep(1)
s.connect((HOST, PORT))
print('OK')

while 1:
    if(face_detect()):
    #if(1):
        #IMAGE_NAME = 'tmp_.jpg' # other image 1
        #IMAGE_NAME = 'tmp.jpg' # other image 2

        filesize = os.path.getsize(IMAGE_NAME)
        s.send(str(filesize).encode())

        f = open(IMAGE_NAME, 'rb')
        while 1:
            data = f.read(2048)
            if data:
                s.send(data)
            else:
                break
        f.close()

        while 1:
            result = s.recv(10)
            if(result == b'1'):
                print('\nGood day Master\n')
                exit()
            else:
                print('\nAuthorization Error.\n')
                exit()

s.close()

 

3. 얼굴 이미지 학습 시 사용한 모델 (CNN)

ㅇ