1 ：原理

计算当出现 1 次眨眼或 1 次张嘴就判断为活人，记录下一帧的人脸图片，和要判定的人员图片进行比对，获取比对后的相似度，进行判断是否是同一个人，为了增加判断的速度，才用 2 帧进行一次活体检测判断。

2 ：代码实现

import face_recognition

from imutils import face_utils

import numpy as np

import dlib

import cv2

import sys

# 初始化眨眼次数

blink_total = 0

# 初始化张嘴次数

mouth_total = 0

# 设置图片存储路径

pic_path = r'images\viode_face.jpg'

# 图片数量

pic_total = 0

# 初始化眨眼的连续帧数以及总的眨眼次数

blink_counter = 0

# 初始化张嘴状态为闭嘴

mouth_status_open = 0

def getFaceEncoding(src):

     image = face_recognition.load_image_file(src)  # 加载人脸图片

     # 获取图片人脸定位 [(top,right,bottom,left )]

     face_locations = face_recognition.face_locations(image)

     img_ = image[face_locations[0][0]:face_locations[0][2], face_locations[0][3]:face_locations[0][1]]

     img_ = cv2.cvtColor(img_, cv2.COLOR_BGR2RGB)

     # display(img_)

     face_encoding = face_recognition.face_encodings(image, face_locations)[0]  # 对人脸图片进行编码

     return face_encoding

def simcos(a, b):

     a = np.array(a)

     b = np.array(b)

     dist = np.linalg.norm(a - b)  # 二范数

     sim = 1.0 / (1.0 + dist)  #

     return sim

# 提供对外比对的接口 返回比对的相似度

def comparison(face_src1, face_src2):

     xl1 = getFaceEncoding(face_src1)

     xl2 = getFaceEncoding(face_src2)

     value = simcos(xl1, xl2)

     print(value)

# 眼长宽比例

def eye_aspect_ratio(eye):

     # (|e1-e5|+|e2-e4|) / (2|e0-e3|)

     A = np.linalg.norm(eye[1] - eye[5])

     B = np.linalg.norm(eye[2] - eye[4])

     C = np.linalg.norm(eye[0] - eye[3])

     ear = (A + B) / (2.0 * C)

     return ear

# 嘴长宽比例

def mouth_aspect_ratio(mouth):

     A = np.linalg.norm(mouth[1] - mouth[7])  # 61, 67

     B = np.linalg.norm(mouth[3] - mouth[5])  # 63, 65

     C = np.linalg.norm(mouth[0] - mouth[4])  # 60, 64

     mar = (A + B) / (2.0 * C)

     return mar

#   进行活体检测（包含眨眼和张嘴）

#  filePath 视频路径

def liveness_detection():

     global blink_total  # 使用 global 声明 blink_total ，在函数中就可以修改全局变量的值

     global mouth_total

     global pic_total

     global blink_counter

     global mouth_status_open

     # 眼长宽比例值

     EAR_THRESH = 0.15

     EAR_CONSEC_FRAMES_MIN = 1

     EAR_CONSEC_FRAMES_MAX = 5  # 当 EAR 小于阈值时，接连多少帧一定发生眨眼动作

     # 嘴长宽比例值

     MAR_THRESH = 0.2

     # 人脸检测器

     detector = dlib.get_frontal_face_detector()

     # 特征点检测器

     predictor = dlib.shape_predictor("model/shape_predictor_68_face_landmarks.dat")

     # 获取左眼的特征点

     (lStart, lEnd) = face_utils.FACIAL_LANDMARKS_IDXS["left_eye"]

     # 获取右眼的特征点

     (rStart, rEnd) = face_utils.FACIAL_LANDMARKS_IDXS["right_eye"]

     # 获取嘴巴特征点

     (mStart, mEnd) = face_utils.FACIAL_LANDMARKS_IDXS["inner_mouth"]

     vs = cv2.VideoCapture(video_path)

     # 总帧数 (frames)

     frames = vs.get(cv2.CAP_PROP_FRAME_COUNT)

     frames_total = int(frames)

     for i in range(frames_total):

         ok, frame = vs.read(i)  # 读取视频流的一帧

         if not ok:

             break

         if frame is not None and i % 2 == 0:

             # 图片转换成灰色（去除色彩干扰，让图片识别更准确）

             gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)

             rects = detector(gray, 0)  # 人脸检测

             # 只能处理一张人脸

             if len(rects) == 1:

                 if pic_total == 0:

                     cv2.imwrite(pic_path, frame)  # 存储为图像 , 保存名为 文件夹名 _ 数字（第几个文件） .jpg

                     cv2.waitKey(1)

                     pic_total += 1

                 shape = predictor(gray, rects[0])  # 保存 68 个特征点坐标的 <class 'dlib.dlib.full_object_detection'> 对象

                 shape = face_utils.shape_to_np(shape)  # 将 shape 转换为 numpy 数组，数组中每个元素为特征点坐标

                 left_eye = shape[lStart:lEnd]  # 取出左眼对应的特征点

                 right_eye = shape[rStart:rEnd]  # 取出右眼对应的特征点

                 left_ear = eye_aspect_ratio(left_eye)  # 计算左眼 EAR

                 right_ear = eye_aspect_ratio(right_eye)  # 计算右眼 EAR

                 ear = (left_ear + right_ear) / 2.0   # 求左右眼 EAR 的均值

                 mouth = shape[mStart:mEnd]   # 取出嘴巴对应的特征点

                 mar =外汇跟单gendan5.com mouth_aspect_ratio(mouth)  # 求嘴巴 mar 的均值

                 # EAR 低于阈值，有可能发生眨眼，眨眼连续帧数加一次

                 if ear < EAR_THRESH:

                     blink_counter += 1

                 # EAR 高于阈值，判断前面连续闭眼帧数，如果在合理范围内，说明发生眨眼

                 else:

                     if EAR_CONSEC_FRAMES_MIN <= blink_counter <= EAR_CONSEC_FRAMES_MAX:

                         blink_total += 1

                     blink_counter = 0

                 # 通过张、闭来判断一次张嘴动作

                 if mar > MAR_THRESH:

                     mouth_status_open = 1

                 else:

                     if mouth_status_open:

                         mouth_total += 1

                     mouth_status_open = 0

             elif len(rects) == 0 and i == 90:

                 print("No face!")

                 break

             elif len(rects) > 1:

                 print("More than one face!")

         # 判断眨眼次数大于 2 、张嘴次数大于 1 则为活体 , 退出循环

         if blink_total >= 1 or mouth_total >= 1:

             break

     cv2.destroyAllWindows()

     vs.release()

# video_path, src = sys.argv[1], sys.argv[2]

video_path = r'video\face13.mp4'      # 输入的 video 文件夹位置

# src = r'C:\Users\666\Desktop\zz5.jpg'

liveness_detection()

print(" 眨眼次数》》 ", blink_total)

print(" 张嘴次数》》 ", mouth_total)

# comparison(pic_path, src)