人脸位置检测

face_locations = face_recognition.face_locations(frame)

检测图像中的人脸

参数说明

frame：按照RGB排列的彩色图像

返回值

face_locations：[(x1,y1,x2,y2)] 表示所有检测到人脸的位置坐标数组，每个数组元素代表一个人脸的位置（左上角和右下角坐标）

人脸数据编码

face_encodings = face_recognition.face_encodings(frame, face_locations)

对检测到的人脸进行编码

参数说明

frame：按照RGB排列的彩色图像
face_locations：检测到的人脸位置

返回值

face_encodings：人脸编码数据形成的数组，数组大小与传入的face_locations一致

人脸比对

face_distances = face_recognition.face_distance(known_face_encodings, face_encoding)

对检测到的人脸进行比对

参数说明

known_face_encodings：已有的人脸编码数组
face_encoding：待比对的人脸编码

返回值

face_distances：和已有人脸编码的距离，距离越小则说明越接近，低于某个阈值则可认为是相同的人脸

使用示例：打开摄像头，识别画面中的人脸

把人脸图片存放在~/Lepi_Data/ros/face_recognizer/known_face目录下，运行下面的程序就能识别到对应的人脸

  1. #!coding:utf-8
  2. import cv2
  3. from PIL import Image, ImageFont, ImageDraw
  4. import os
  5. import time
  6. import face_recognition
  7. import numpy as np
  9. Key_Esc = 27
  11. defaultFont = ImageFont.truetype(
  12.     '/usr/share/fonts/truetype/wqy/wqy-zenhei.ttc', 48)
  15. def putText(frame, text, pos, color, font=defaultFont):
  16.     """
  17.     将文本显示在图片上
  18.     Keyword arguments:
  19.     frame: image 原图
  20.     text: str 想要显示的文本
  21.     pos: (x,y) 指定显示的初始坐标(左上角顶点)
  22.     color: [r,g,b] 指定文本颜色的r、g、b值
  23.     Returns:
  24.     cv_img: image 叠加了文本的新图片(不改变原图)
  25.     """
  26.     if hasattr(text, "decode"):
  27.         text = text.decode('utf-8')
  28.     pil_image = Image.fromarray(cv2.cvtColor(frame, cv2.COLOR_BGR2RGB))
  29.     draw = ImageDraw.Draw(pil_image)
  30.     draw.text(pos, text, font=font, fill=color)
  31.     cv_img = cv2.cvtColor(np.asarray(pil_image), cv2.COLOR_RGB2BGR)
  32.     return cv_img
  35. class FaceRecognizer(object):
  36.     """
  37.     FaceRecognizer类, 用来检测和识别人脸
  38.     Attributes:
  39.         data_dir: str 标记人脸图片的保存目录
  40.         scale: int 缩放倍数，加快检测速度但会降低检测精度
  41.         font: ImageFont 中文字体
  42.         known_faces: dict 存放已标记人脸的字典
  43.         threshold: float 检测阈值，小于该值才会进行相似度比较
  44.     """
  46.     def __init__(self, scale=5, threshold=0.45, fontSize=18):
  47.         super(FaceRecognizer, self).__init__()
  48.         self.data_dir = os.path.expanduser(
  49.             '~') + "/Lepi_Data/ros/face_recognizer/known_face"
  50.         self.scale = scale
  51.         # self.font = ImageFont.truetype(
  52.         #     '/usr/share/fonts/truetype/droid/DroidSansFallbackFull.ttf', fontSize)
  53.         self.known_faces = {}
  54.         self.threshold = threshold
  55.         self.face_locations = []
  56.         self.face_names = []
  57.         self.face_data = []
  58.         self.add_label_success = False
  59.         self.load_faces()
  61.     def detect(self, frame, scale=None):
  62.         """
  63.         人脸检测函数
  64.         Keyword arguments:
  65.         frame: image 原图
  66.         scale: float 缩放倍数
  67.         Returns:
  68.         face_locations: [(x1,y1,x2,y2)] 表示所有检测到人脸的位置坐标数组，每个数组元素代表一个人脸的位置
  69.         """
  70.         if scale is None:
  71.             scale = self.scale
  72.         # Resize frame of video to 1/self.scale size for faster face recognition processing
  73.         small_frame = cv2.resize(frame, (0, 0), fx=1.0/scale, fy=1.0/scale)
  74.         # Convert the image from BGR color (which OpenCV uses) to RGB color (which face_recognition uses)
  75.         rgb_small_frame = cv2.cvtColor(small_frame, cv2.COLOR_BGR2RGB)
  76.         self.face_locations = face_recognition.face_locations(rgb_small_frame)
  77.         self.face_data = self.getFaceData(0)
  78.         return self.rect_faces(frame, self.face_locations)
  80.     def recognize(self, frame, scale=None):
  81.         """
  82.         人脸识别函数
  83.         Keyword arguments:
  84.         frame: image 原图
  85.         scale: float 缩放倍数
  86.         Returns:
  87.         face_locations: [(x1,y1,x2,y2)] 表示所有检测到人脸的位置坐标数组，每个数组元素代表一个人脸的位置
  88.         face_names: ['未知'] 对应每个人脸的标签，未检测到用'未知'表示
  89.         """
  90.         if scale is None:
  91.             scale = self.scale
  92.         small_frame = cv2.resize(frame, (0, 0), fx=1.0/scale, fy=1.0/scale)
  93.         rgb_small_frame = cv2.cvtColor(small_frame, cv2.COLOR_BGR2RGB)
  94.         face_locations = face_recognition.face_locations(rgb_small_frame)
  95.         start = time.time()
  96.         face_encodings = face_recognition.face_encodings(
  97.             rgb_small_frame, face_locations)
  98.         face_names = []
  99.         # python 3.9 默认生成dict_values 手动转成list进行比较
  100.         known_face_encodings = list(self.known_faces.values())
  101.         known_face_names = list(self.known_faces.keys())
  102.         for face_encoding in face_encodings:
  103.             name = "未知"
  104.             if len(known_face_encodings) > 0:
  105.                 # matches = face_recognition.compare_faces(known_face_encodings, face_encoding)
  106.                 # print(known_face_encodings, face_encoding)
  107.                 face_distances = face_recognition.face_distance(
  108.                     known_face_encodings, face_encoding)
  109.                 best_match_index = np.argmin(face_distances)
  110.                 # print(matches)
  111.                 print(face_distances)
  112.                 if face_distances[best_match_index] < self.threshold:
  113.                     name = known_face_names[best_match_index]
  114.             face_names.append(name)
  115.         end = time.time()
  116.         print("recognized %d faces in %.2f ms" %
  117.               (len(face_locations), (end - start)*1000))
  118.         self.face_locations = face_locations
  119.         self.face_names = face_names
  120.         self.face_data = self.getFaceData(0)
  121.         return self.label_faces(frame, self.face_locations, self.face_names)
  122.         # return face_locations, face_names
  124.     def getFaceData(self, index):
  125.         if len(self.face_locations) > index:
  126.             (top, right, bottom, left) = self.face_locations[index]
  127.             top *= self.scale
  128.             right *= self.scale
  129.             bottom *= self.scale
  130.             left *= self.scale
  131.             return [int((right+left)/2), int((top+bottom)/2), int(right-left), int(bottom-top)]
  132.         else:
  133.             return [0, 0, 0, 0]
  135.     def detectedFaceLabel(self, name):
  136.         if name in self.face_names:
  137.             index = self.face_names.index(name)
  138.             self.face_data = self.getFaceData(index)
  139.             return True
  140.         else:
  141.             self.face_data = []
  142.             return False
  144.     def rect_faces(self, frame, face_locations, scale=None):
  145.         """
  146.         把检测到的人脸用矩形框出来
  147.         Keyword arguments:
  148.         frame: image 原图
  149.         face_locations: [(x1,y1,x2,y2)] 表示所有检测到人脸的位置坐标数组，每个数组元素代表一个人脸的位置
  150.         scale: float 缩放倍数
  151.         Returns:
  152.         frame: image 框出了人脸的新图像(改变原图)
  153.         """
  154.         if scale is None:
  155.             scale = self.scale
  156.         # Display the results
  157.         for (top, right, bottom, left) in face_locations:
  158.             # Scale back up face locations since the frame we detected in was scaled to 1/4 size
  159.             top *= scale
  160.             right *= scale
  161.             bottom *= scale
  162.             left *= scale
  163.             # Draw a box around the face
  164.             cv2.rectangle(frame, (left, top), (right, bottom), (0, 0, 255), 4)
  165.             # Draw a label with a name below the face
  166.         return frame
  168.     def label_faces(self, frame, face_locations, face_names, scale=None):
  169.         """
  170.         给识别到的人脸添加标签
  171.         Keyword arguments:
  172.         frame: image 原图
  173.         face_locations: [(x1,y1,x2,y2)] 表示所有检测到人脸的位置坐标数组，每个数组元素代表一个人脸的位置
  174.         scale: float 缩放倍数，需要和检测时的一致，否则比例会出错
  175.         Returns:
  176.         frame: image 标记了人脸的新图像(不改变原图)
  177.         """
  178.         if scale is None:
  179.             scale = self.scale
  180.         frame = self.rect_faces(frame, face_locations, scale)
  181.         # Display the results
  182.         for (top, right, bottom, left), name in zip(face_locations, face_names):
  183.             # Scale back up face locations since the frame we detected in was scaled to 1/4 size
  184.             top *= scale
  185.             right *= scale
  186.             bottom *= scale
  187.             left *= scale
  188.             # Draw a box around the face
  189.             # cv2.rectangle(frame, (left, top), (right, bottom), (0, 0, 255), 2)
  190.             # Draw a label with a name below the face
  191.             color = (0, 0, 255)
  192.             frame = putText(frame, name, (left + 6, bottom ), color)
  193.         return frame
  195.     def load_faces(self):
  196.         """
  197.         加载本地的人脸标签(默认存放在data_dir下)
  198.         Keyword arguments:
  199.         无
  200.         Returns:
  201.         无
  202.         """
  203.         if not os.path.exists(self.data_dir):
  204.             os.makedirs(self.data_dir)
  205.         files = os.listdir(self.data_dir)
  206.         for file in files:
  207.             try:
  208.                 name = file.split('.')[0]
  209.                 file_path = os.path.join(self.data_dir, file)
  210.                 print(self.add_face_label(cv2.imread(file_path), name, scale=1))
  211.             except Exception as e:
  212.                 print(e)
  214.     def add_face_label(self, frame, name, scale=None, save=False):
  215.         """
  216.         动态添加人脸标签
  217.         Keyword arguments:
  218.         frame: image 原图
  219.         name: str 标签名称
  220.         scale: float 缩放倍数
  221.         save: bool 是否保存，True则将保存图片至本地标签目录，每次启动会重新读取，False只在本次运行生效
  222.         Returns:
  223.         无
  224.         """
  225.         if scale is None:
  226.             scale = self.scale
  227.         frame = cv2.resize(frame, (0, 0), fx=1.0/scale, fy=1.0/scale)
  228.         rgb_frame = frame[:, :, ::-1]
  229.         start = time.time()
  230.         self.detect(frame, scale=1)
  231.         face_locations = self.face_locations
  232.         end = time.time()
  233.         print("Found %d faces in %.2f ms" %
  234.               (len(face_locations), (end - start)*1000))
  236.         if len(face_locations) == 1:
  237.             face_encoding = face_recognition.face_encodings(
  238.                 frame, face_locations)[0]
  239.             self.known_faces[name] = face_encoding
  240.             if save:
  241.                 file_path = os.path.join(self.data_dir, name+'.png')
  242.                 print(file_path)
  243.                 cv2.imwrite(file_path, frame)
  244.             print('成功添加"%s"的标记' % (name))
  245.             self.add_label_success = True
  246.             return '标记成功'
  247.         elif len(face_locations) > 1:
  248.             self.add_label_success = False
  249.             return '标记"%s"失败，检测到多余人脸' % (name)
  250.         else:
  251.             self.add_label_success = False
  252.             return '标记"%s"失败，未检测到人脸' % (name)
  254.     def remove_face_label(self, name):
  255.         """
  256.         删除人脸标签
  257.         Keyword arguments:
  258.         name: str 标签名称
  259.         Returns:
  260.         无
  261.         """
  262.         if self.known_faces.__contains__(name):
  263.             self.known_faces.pop(name)
  264.         try:
  265.             os.system('rm '+os.path.join(self.data_dir, name+".*"))
  266.             return "已删除"
  267.         except Exception as e:
  268.             print(e)
  269.             return "删除出错"
  272. def test_recognize():
  273.     """
  274.     测试函数，打开摄像头实时检测人脸, 按q退出
  275.     Keyword arguments:
  276.     无
  277.     Returns:
  278.     无
  279.     """
  280.     fr = FaceRecognizer()
  281.     cap = cv2.VideoCapture(0)
  282.     while True:
  283.         # Grab a single frame of video
  284.         ret, frame = cap.read()
  285.         start = time.time()
  286.         nframe = fr.recognize(frame)
  287.         end = time.time()
  288.         print("labeled %d faces in %.2f ms" %
  289.               (len(fr.face_locations), (end - start)*1000))
  290.         cv2.imshow("images", np.rot90(cv2.resize(nframe, (320, 240))))
  291.         c = cv2.waitKey(1)
  292.         if c == Key_Esc:
  293.             break
  294.     cap.release()
  295.     cv2.destroyAllWindows()
  298. if __name__ == '__main__':
  299.     test_recognize()