配合式活体检测的python实现
前段时间发布的活体检测相关的调研文章中提到目前常用的检测方法有配合式活体检测、静默活体检测以及双目活体防伪检测. 本文将实现配合式的活体检测, 主要从常见的脸部的眨眼、 张嘴、 摇头、 点头四个操作来进行活体检测.
首先框定人脸, 然后通过寻找人脸中的68个关键点, 在这四个操作中随机选取3个操作, 根据关键点的变化来观察是否有完成上述操作过程.
特征检测器shape_predictor_68_face_landmarks.dat下载链接
具体实现代码如下:
#!/usr/bin/env python
# -*- coding: utf-8 -*-
# @Time : 2020/3/24 2:28 下午
# @Author : sarah feng
# @File : operate—detect.py
from scipy.spatial import distance as dist
from imutils.video import VideoStream
from imutils import face_utils
from threading import Thread
import numpy as np
import argparse
import imutils
import time
import dlib
import random
import cv2
import time
class OperateDetect(object):
def __init__(self):
self.weight = "./shape_predictor_68_face_landmarks.dat"
self.EYE_AR_THRESH = 0.25 # 眼睛对应的长宽比低于该阈值算一次眨眼
self.CONSEC_FRAMES = 5 # 这个值被设置为 3，表明眼睛长宽比小于3时，接着三个连续的帧一定发生眨眼动作
self.COUNTER = 0 # 眼图长宽比小于EYE_AR_THRESH的连续帧的总数
self.TOTAL = 0 # 脚本运行时发生的眨眼的总次数
self.MOUTH_AR_THRESH = 0.8
self.Nod_threshold = 0.03
self.shake_threshold = 0.03
self.detector = dlib.get_frontal_face_detector()
self.predictor = dlib.shape_predictor(self.weight)
self.mStart, self.mEnd = 49, 68 # 嘴部对应的索引
self.nStart, self.nEnd = 32, 36 #鼻子对应的索引
self.lStart, self.lEnd = face_utils.FACIAL_LANDMARKS_IDXS["left_eye"] #左眼对应的的索引
self.rStart, self.rEnd = face_utils.FACIAL_LANDMARKS_IDXS["right_eye"] #右眼对应的索引
self.compare_point = [0, 0] # 刚开始的时候设置鼻子的中点在【0， 0】
def eye_aspect_ratio(self, eye):
'''
A，B是计算两组垂直眼睛标志之间的距离，而C是计算水平眼睛标志之间的距离
# compute the euclidean distances between the two sets of
# vertical eye landmarks (x, y)-coordinates
:param eye:两个眼睛在图片中的像素坐标
:return:返回眼睛的长宽比
'''
A = dist.euclidean(eye[1], eye[5])
B = dist.euclidean(eye[2], eye[4])
C = dist.euclidean(eye[0], eye[3])
ear = (A + B) / (2.0 * C) # 将分子和分母相结合，得出最终的眼睛纵横比。然后将眼图长宽比返回给调用函数
return ear
def mouth_aspect_ratio(self, mouth):
# 计算两组垂直方向的欧式距离
A = dist.euclidean(mouth[2], mouth[10]) # 51, 59
B = dist.euclidean(mouth[4], mouth[8]) # 53, 57
# 计算水平方向的距离
C = dist.euclidean(mouth[0], mouth[6]) # 49, 55
mar = (A + B) / (2.0 * C)
return mar
def center_point(self, nose):
return nose.mean(axis=0)
def nod_aspect_ratio(self, size, pre_point, now_point):
return abs(float((pre_point[1] - now_point[1]) / (size[0] / 2)))
def shake_aspect_ratio(self, size, pre_point, now_point):
return abs(float((pre_point[0] - now_point[0]) / (size[1] / 2)))
def blinks_detect(self, shape):
leftEye = shape[self.lStart:self.lEnd]
rightEye = shape[self.rStart:self.rEnd]
leftEAR = self.eye_aspect_ratio(leftEye)
rightEAR = self.eye_aspect_ratio(rightEye)
# average the eye aspect ratio together for both eyes
ear = (leftEAR + rightEAR) / 2.0
# 计算左右眼的凸包，并可视化
leftEyeHull = cv2.convexHull(leftEye)
rightEyeHull = cv2.convexHull(rightEye)
return ear, leftEyeHull, rightEyeHull
def mouth_detect(self, shape):
mouth = shape[self.mStart:self.mEnd]
mouthMAR = self.mouth_aspect_ratio(mouth)
mar = mouthMAR
mouthHull = cv2.convexHull(mouth)
return mar, mouthHull
def nod_shark(self, size, shape):
nose = shape[self.nStart:self.nEnd]
nose_center = self.center_point(nose)
nod_value, shake_value = 0, 0
noseHull = cv2.convexHull(nose)
if self.compare_point[0] != 0:
nod_value = self.nod_aspect_ratio(size, nose_center, self.compare_point)
shake_value = self.shake_aspect_ratio(size, nose_center, self.compare_point)
self.compare_point = nose_center
return nod_value, shake_value, noseHull
def in_area(self, key_points, circle_center, radius):
result = True
left_eyebrow= key_points[18:22]
right_eyebrow = key_points[18:22]
mouth = key_points[49:68]
left_eyebrow_center = self.center_point(left_eyebrow)
right_eyebrow_center = self.center_point(right_eyebrow)
mouth_center = self.center_point(mouth)
# 首先判断三个点的是否在圆内
if np.sqrt(np.sum((circle_center - left_eyebrow_center) ** 2)) - radius > 0:
result = False
if np.sqrt(np.sum((circle_center - right_eyebrow_center) ** 2)) - radius > 0:
result = False
if np.sqrt(np.sum((circle_center - mouth_center) ** 2)) - radius > 0:
result = False
if left_eyebrow_center[0] > circle_center[0] or left_eyebrow_center[1] > circle_center[1]:
result = False
if right_eyebrow_center[0] < circle_center[0] *0.9 or left_eyebrow_center[1] > circle_center[1]:
result = False
if mouth_center[1] self.MOUTH_AR_THRESH:
action_type[1] = 1
if nod_value > self.Nod_threshold:
action_type[2] = 1
if shake_value > self.shake_threshold:
action_type[3] = 1
return action_type
def detect(self, frame):
img = imutils.resize(frame, width=640)
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
rects = self.detector(gray, 0)
size = frame.shape
if len(rects) != 0:
for rect in rects: # 遍历帧中的每个面，然后对其中的每个面应用面部标志检测
# shape确定面部区域的面部标志，接着将这些（x，y）坐标转换成NumPy阵列
shape = self.predictor(img, rect)
shape = face_utils.shape_to_np(shape)
in_circle = self.in_area(shape, np.array([320, 180]), 90)
ear, leftEyeHull, rightEyeHull = self.blinks_detect(shape)
mar, mouthHull = self.mouth_detect(shape)
nod_value, shake_value, noseHull = self.nod_shark(size, shape)
act_type = self.action_judgment((ear, mar, nod_value, shake_value))
return in_circle, act_type, leftEyeHull, rightEyeHull, mouthHull, noseHull
def main():
vs = VideoStream(src=0).start()
live_detect = OperateDetect()
# 每次运行的时候做三中方式的检测， 顺序打乱
detect_type = [0, 1, 2, 3]
random.shuffle(detect_type)
detect_type = detect_type[0:-1]
frame = vs.read()
frame = imutils.resize(frame, width=640)
size = frame.shape
activate_judge = np.array([0, 0, 0, 0])
first_frame_type = True
while True:
frame = vs.read()
frame = imutils.resize(frame, width=640)
cv2.circle(frame, (int(size[1] / 2), int(size[0] / 2)), int(min(size[0] / 2, size[1] / 2) * 0.5), (0, 0, 255))
detect_result = live_detect.detect(frame)
operate_step = True
if len(detect_type) ==0:
break
if detect_result:
if(len(detect_result)) ==6:
is_align, act_info, leftEye, rightEye, mouth_point, nose_point = detect_result
detect_num = detect_type[0]
if first_frame_type:
if not is_align:
operate_step = False
if operate_step:
if detect_num == 0:
cv2.putText(frame, "Trouble blinking", (200, 100),
cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 0, 255), 2)
if detect_num == 1: # 张嘴操作
cv2.putText(frame, "Open Mouth", (200, 100),
cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 0, 255), 2)
if detect_num == 2: # 点头
cv2.putText(frame, "Nod", (200, 100),
cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 0, 255), 2)
if detect_num == 3: # 摇头操作
cv2.putText(frame, "Shake head", (200, 100),
cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 0, 255), 2)
if is_align:
first_frame_type = False
if act_info[detect_num] == 1:
activate_judge = activate_judge + act_info
if activate_judge[detect_num] > 3:
cv2.putText(frame, "Single test completed!", (30, 60),
cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 0, 255), 2)
print("detect_type[0]", detect_type[0])
detect_type.remove(detect_type[0])
first_frame_type = True
else:
activate_judge = np.array([0, 0, 0, 0])
cv2.drawContours(frame, [leftEye], -1, (0, 255, 0), 1)
cv2.drawContours(frame, [rightEye], -1, (0, 255, 0), 1)
cv2.drawContours(frame, [mouth_point], -1, (0, 255, 0), 1)
cv2.drawContours(frame, [nose_point], -1, (0, 255, 0), 1)
else:
cv2.putText(frame, "Trouble align your face in circle", (200, 100),
cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 0, 255), 2)
cv2.imshow("Frame", frame)
key = cv2.waitKey(1) & 0xFF
if len(detect_type) == 0:
break
cv2.destroyAllWindows()
vs.stop()
if __name__ == '__main__':
main()