基于dlib的人脸68特征点提取(眨眼张嘴检测)python版本

引言

(1) 利用 Dlib 官方训练好的模型 “shape_predictor_68_face_landmarks.dat” 进行 68 个点标定&＃xff1b;
(2)利用 OpenCv 进行图像化处理&＃xff0c;在人脸上画出 68 个特征点&＃xff0c;并标明特征点的序号实现的 68 个特征点&＃xff1b;
(3)根据得到的点位进行简单的活体检测

开发环境和库

1.Window10
2.Anaconda3
3.Python 3.6
4.dlib 19.8.1
5.opencv-python 4.2.0.32 &＃xff08;openCV只是用来画图&＃xff09;

PS:dlib 的pip install 可能需要加上版本号才能成功下载安装
pip install dlib&＃61;&＃61;19.8.1

# -*- coding: UTF-8 -*-
import dlib
from skimage import io
from scipy.spatial import distance as dist
import matplotlib.pyplot as plt
import numpy as np
import cv2


流程设计

1. 68 点提取&＃xff1a;

借助 Dlib 官方的 Demo: face_landmark_detection.py&＃xff0c;可以得到脸部 68 个特征点的坐标&＃xff1b;

2. OpenCv 绘图&＃xff1a;

使用 opencv 中 画圆函数 cv2.circle() 和 画字符函数 cv2.putText() &＃xff1b;
静态图片的检测

张嘴和闭眼的检测

参考&＃xff1a;Eye blink detection with OpenCV, Python, and dlib
公式很简单&＃xff0c;就是欧式距离公式&＃xff0c;设定好纵横比(EAR)。

def eye_aspect_ratio(eye):# 垂直眼标志&＃xff08;X&＃xff0c;Y&＃xff09;坐标A &＃61; dist.euclidean(eye[1],eye[5])#计算两个集合之间的欧式距离B &＃61; dist.euclidean(eye[2],eye[4])#水平C &＃61; dist.euclidean(eye[0], eye[3])ear &＃61; (A&＃43;B) / (2.0 * C)return ear


def mouth_aspect_ratio(mouth):#垂直点位A &＃61; np.linalg.norm(mouth[2] - mouth[9])B &＃61; np.linalg.norm(mouth[4] - mouth[7])C &＃61; np.linalg.norm(mouth[0] - mouth[6])mar &＃61; (A &＃43; B) / (2.0 * C)return mar


还有可调试的常量

EYE_AR_THRESH &＃61; 0.2
EYE_AR_CONSEC_FRAMES &＃61; 2
MAR_THRESH &＃61; 0.5
MOUTH_AR_CONSEC_FRAMES &＃61; 3
COUNT &＃61; 0
TOTAL &＃61; 0
mCOUNT &＃61; 0
mTOTAL &＃61; 0
r_eye_ear &＃61; 0
l_eye_ear &＃61; 0


然后我们就可以把视频流的定帧数据输入到我们的函数里面解析&＃xff0c;算出各个指标来判断是否触发张嘴闭眼。

cap &＃61; cv2.VideoCapture(0) #创建一个对象
cap.set(3,720)
isOpen &＃61; cap.isOpened() #检测是否开启
while isOpen:#cap.read()#返回两个值&＃xff1a;#一个布尔值 true/false&＃xff0c;用来判断读取视频是否成功/是否到视频末尾#图像对象&＃xff0c;图像的三维矩阵flag,im_rd &＃61; cap.read()# 每帧数据延时 1ms&＃xff0c;延时为 0 读取的是静态帧k &＃61; cv2.waitKey(1)img_gray &＃61; cv2.cvtColor(im_rd, cv2.COLOR_RGB2GRAY)faces &＃61; detector(img_gray, 0)font &＃61; cv2.FONT_HERSHEY_SIMPLEX# 标点if len(faces) !&＃61; 0:# 检测到人脸for i in range(len(faces)):landmarks &＃61; np.matrix([[p.x, p.y] for p in landmark_predictor(im_rd, faces[i]).parts()])left_eye &＃61; landmarks[42:48]right_eye &＃61; landmarks[36:41]mouth_points &＃61; landmarks[48:68]#print(left_eye)l_eye_ear &＃61; eye_aspect_ratio(left_eye)r_eye_ear &＃61; eye_aspect_ratio(left_eye)t_ear &＃61; (l_eye_ear &＃43;r_eye_ear) / 2.0mouth_ear &＃61; mouth_aspect_ratio(mouth_points)if t_ear < 0.2:COUNT &＃43;&＃61; 1else:if COUNT >&＃61; EYE_AR_CONSEC_FRAMES:TOTAL &＃43;&＃61; 1COUNT &＃61; 0if mouth_ear > MAR_THRESH:mCOUNT &＃43;&＃61; 1else:if mCOUNT >&＃61; MOUTH_AR_CONSEC_FRAMES:mTOTAL &＃43;&＃61; 1mCOUNT &＃61; 0for idx,point in enumerate(landmarks):# 68点打印pos &＃61; (point[0, 0], point[0, 1])#利用 cv2.circle 给每个特征点画一个圈&＃xff0c;共 68 个cv2.circle(im_rd, pos, 2, color&＃61;(139, 0, 0))# 利用 cv2.putText 输出 1-68#cv2.putText(im_rd, str(idx &＃43; 1), pos, font, 0.2, (187, 255, 255), 1, cv2.LINE_AA)cv2.putText(im_rd, "faces: " &＃43; str(len(faces)), (20, 20), font, 1, (255,20,147), 1, cv2.LINE_AA)#cv2.putText(im_rd, "L_Eye_Ear: " &＃43; str(l_eye_ear), (20, 60), font, 1, (0, 0, 0), 1, cv2.LINE_AA)#cv2.putText(im_rd, "R_Eye_Ear: " &＃43; str(r_eye_ear), (20, 90), font, 1, (0, 0, 0), 1, cv2.LINE_AA)cv2.putText(im_rd, "T_Ear: " &＃43; str(t_ear), (20, 40), font, 1, (255,20,147), 1, cv2.LINE_AA)cv2.putText(im_rd, "BlinkNum: " &＃43; str(TOTAL), (20, 80), font, 1, (255,20,147), 1, cv2.LINE_AA)cv2.putText(im_rd, "openMouth: " &＃43; str(mTOTAL), (20, 100), font, 1, (255,20,147), 1, cv2.LINE_AA)else:cv2.putText(im_rd, "no face", (20, 50), font, 1, (0, 0, 0), 1, cv2.LINE_AA)cv2.putText(im_rd, "Press &＃39;q&＃39;: Quit", (20, 450),cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0,0,255),)cv2.namedWindow("camera", 0)cv2.imshow("camera", im_rd)if cv2.waitKey(1) & 0xFF &＃61;&＃61; ord(&＃39;q&＃39;):break
cap.release()
cv2.destroyAllWindows()