OpenCV/DLib人脸检测

在python环境下实现人脸检测有多种方式，最简单易用的是使用opencv或者dlib，两者都提供了经典方法和深度学习方法进行人脸检测：

1. opencv haar cascade检测：haarcascade_frontalface_alt.xml
2. opencv dnn模块加载预训练模型：res10_300x300_ssd_iter_140000_fp16.caffemodel+deploy.prototxt，或者TensorFlow的量化版本opencv_face_detector_uint8.pb+opencv_face_detector.pbtxt
3. dlib mmod
4. dlib hog
   关于opencv中dnn人脸检测预训练模型，可以查看opencv自带的例子（需要下载源码解压）"F:\opencv\sources\samples\dnn\face_detector\download_weights.py"

1. opencv dnn ssd300

下面的两段代码都可以实现dnn模块加载深度学习模型，第二段代码使用cv2.dnn_DetectionModel接口比较简洁。

import numpy as np
import argparse
import cv2

args = {
     'prototxt':r"F:\opencv\sources\samples\dnn\face_detector\deploy.prototxt",
        'model':r"F:\opencv\sources\samples\dnn\face_detector\res10_300x300_ssd_iter_140000_fp16.caffemodel",
        'image':r"C:\Users\admin\Pictures\faces.png",
        'confidence':0.5}
# load our serialized model from disk
print("[INFO] loading model...")
net = cv2.dnn.readNetFromCaffe(args["prototxt"], args["model"])

# load the input image and construct an input blob for the image
# by resizing to a fixed 300x300 pixels and then normalizing it
image = cv2.imread(args["image"])
(h, w) = image.shape[:2]
blob = cv2.dnn.blobFromImage(cv2.resize(image, (300, 300)), 
                             scalefactor=1.0,
                             size=(300, 300), 
                             mean=(104.0, 177.0, 123.0),
                             swapRB=True
                             )

# pass the blob through the network and obtain the detections and
# predictions
print("[INFO] computing object detections...")
net.setInput(blob)
detections = net.forward()

# loop over the detections
for i in range(0, detections.shape[2]):
	# extract the confidence (i.e., probability) associated with the
	# prediction
	confidence = detections[0, 0, i, 2]

	# filter out weak detections by ensuring the `confidence` is
	# greater than the minimum confidence
	if confidence > args["confidence"]:
		# compute the (x, y)-coordinates of the bounding box for the
		# object
		box = detections[0, 0, i, 3:7] * np.array([w, h, w, h])
		(startX, startY, endX, endY) = box.astype("int")
 
		# draw the bounding box of the face along with the associated
		# probability
		text = "{:.2f}%".format(confidence * 100)
		y = startY - 10 if startY - 10 > 10 else startY + 10
		cv2.rectangle(image, (startX, startY), (endX, endY),
			(0, 0, 255), 2)
		cv2.putText(image, text, (startX, y),
			cv2.FONT_HERSHEY_SIMPLEX, 0.45, (0, 0, 255), 2)

# show the output image
cv2.imshow("Output", image)
cv2.waitKey(0)
cv2.destroyAllWindows()

#%% 
# 加载人脸检测模型
model_path = r"F:\opencv\sources\samples\dnn\face_detector\opencv_face_detector_uint8.pb"
pbtxt_path=r"F:\opencv\sources\samples\dnn\face_detector\opencv_face_detector.pbtxt"
# net = cv2.dnn_DetectionModel(args["model"], args["prototxt"])
net = cv2.dnn_DetectionModel(model_path, pbtxt_path)
# net.setPreferableBackend(cv2.dnn.DNN_BACKEND_CUDA) 
# net.setPreferableTarget(cv2.dnn.DNN_TARGET_CUDA)
net.setInputSize(300,300) # 设置网络输入尺寸
net.setInputScale(1.0)   # 注意这里没有归一化
net.setInputSwapRB(True)

frame=cv2.imread(args["image"])

import time
s = time.time()
for _ in range(50):
    classes, confs, boxes = net.detect(frame, 0.5, 0.5)
print('平均推理时间:', (time.time() - s) / 50.0)
classes, confs, boxes = net.detect(frame, 0.5, 0.5)
names = ['_background_', 'face']
for id, conf, box in zip(classes.flatten(), confs.flatten(), boxes):
    label = '{}, {:.2f}'.format(names[id], conf)
    # print(label)
    labelsize, baseLine= cv2.getTextSize(label, cv2.FONT_HERSHEY_SIMPLEX,0.5,1)
    left, top, width, height = box
    top = max(top, labelsize[1])
    cv2.rectangle(frame, box, color=(0, 255, 0), thickness=3)
    cv2.rectangle(frame, (left, top-labelsize[1]),
                 (left+labelsize[0], top+baseLine),(255, 255, 255), cv2.FILLED)
    cv2.putText (frame, label,(left, top), cv2.FONT_HERSHEY_SIMPLEX, 0.5,(0,0, 0))

cv2.imshow('frame', frame)
cv2.waitKey(0)
cv2.destroyAllWindows()

2.Opencv Haar Cascade

取消相关注释之后可以检测眼睛。

#%% haar
def detectAndDisplay(frame):
    frame_gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
    frame_gray = cv2.equalizeHist(frame_gray)
    #-- Detect faces
    faces = face_cascade.detectMultiScale(frame_gray)
    for (x,y,w,h) in faces:
        center = (x + w//2, y + h//2)
        frame = cv2.ellipse(frame, center, (w//2, h//2), 0, 0, 360, (255, 0, 255), 4)
        # faceROI = frame_gray[y:y+h,x:x+w]
        # #-- In each face, detect eyes
        # eyes = eyes_cascade.detectMultiScale(faceROI)
        # for (x2,y2,w2,h2) in eyes:
        #     eye_center = (x + x2 + w2//2, y + y2 + h2//2)
        #     radius = int(round((w2 + h2)*0.25))
        #     frame = cv2.circle(frame, eye_center, radius, (255, 0, 0 ), 4)
    return frame

face_cascade = cv2.CascadeClassifier()
eyes_cascade = cv2.CascadeClassifier()
#-- 1. Load the cascades
if not face_cascade.load(r"F:\opencv-4.4.0\data\haarcascades\haarcascade_frontalface_alt.xml"):
    print('--(!)Error loading face cascade')
    exit(0)
if not eyes_cascade.load(r"F:\opencv-4.4.0\data\haarcascades\haarcascade_eye.xml"):
    print('--(!)Error loading eyes cascade')
    exit(0)
frame=cv2.imread(args["image"])

import time
s = time.time()
for _ in range(50):
    _ = detectAndDisplay(frame)
print('平均推理时间:', (time.time() - s) / 50.0)
frame = detectAndDisplay(frame)

cv2.imshow('frame', frame)
cv2.waitKey(0)
cv2.destroyAllWindows()

3.DLIB CNN MMOD模型

预训练模型可以从http://dlib.net/files/mmod_human_face_detector.dat.bz2下载。

#%% dlib
# get the mmod_human_face_detector.dat file from: 
    # http://dlib.net/files/mmod_human_face_detector.dat.bz2
f=r'E:\Applications\WPy64-3741\python-3.7.4.amd64\Lib\site-packages\face_recognition_models\models\mmod_human_face_detector.dat'
cnn_face_detector = dlib.cnn_face_detection_model_v1(f)
win = dlib.image_window()
img = dlib.load_rgb_image(args['image'])
import time
s = time.time()
for _ in range(10):
    _ = cnn_face_detector(img, 1)
print('平均推理时间:', (time.time() - s) / 10.0)
dets = cnn_face_detector(img, 1)
print("Number of faces detected: {}".format(len(dets)))
for i, d in enumerate(dets):
    print("Detection {}: Left: {} Top: {} Right: {} Bottom: {} Confidence: {}".format(
        i, d.rect.left(), d.rect.top(), d.rect.right(), d.rect.bottom(), d.confidence))

rects = dlib.rectangles()
rects.extend([d.rect for d in dets])

win.clear_overlay()
win.set_image(img)
win.add_overlay(rects)
dlib.hit_enter_to_continue()

4. DLIB HOG模型

#%% dlib
import dlib
import matplotlib.pyplot as plt
def rect_to_bb(rect):
	# take a bounding predicted by dlib and convert it
	# to the format (x, y, w, h) as we would normally do
	# with OpenCV
	x = rect.left()
	y = rect.top()
	w = rect.right() - x
	h = rect.bottom() - y
	# return a tuple of (x, y, w, h)
	return (x, y, w, h)
# gray = cv2.imread(args['image'],1)
from PIL import Image
img = np.array(Image.open(args['image']).convert('RGB'))

face_detect = dlib.get_frontal_face_detector()

import time
s = time.time()
for _ in range(50):
    _ = face_detect(img, 1)
print('平均推理时间:', (time.time() - s) / 50.0)
rects = face_detect(img, 1)

for (i, rect) in enumerate(rects):
    (x, y, w, h) = rect_to_bb(rect)
    cv2.rectangle(img, (x, y), (x + w, y + h), (255, 255, 255), 3)

plt.figure(figsize=(12,8))
plt.imshow(img, cmap='gray')
plt.show()

测试图片为：

最终这4种模型的检测对比结果如下图所示。

检测速度如下(CPU环境，配置为Intel®Xeon® CPU E5-2650 v2@ 2.60GHz 2.60 GHz）：

1. opencv dnn: 0.042s
2. opencv haar: 0.118s
3. dlib hog: 0.249s
4. dlib mmod：35.6s

结合检测速度和实际检测效果，opencv的传统人脸检测方法仍然具有较大优势，对于图中的小尺寸人脸haar模型可以正确识别，其他模型都无法检测出来。另外，dlib的两个模型效果都差强人意，特别是cnn版本的人脸检测模型，检测速度实在太慢，精度还一般。