人脸关键点检测：主流算法与开源工具汇总

开源库有dlib

安装：https://pypi.org/simple/dlib/ ， 下载  后pip安装

模型下载：Index of /files

1、dlib5点

左右眼角及人中5点。

import dlibimport numpy as npimport cv2import time predictor_path = '../model/shape_predictor_5_face_landmarks.dat'detector = dlib.get_frontal_face_detector()predictor = dlib.shape_predictor(predictor_path) start_time = time.time() frame = cv2.imread("../image/ab.jpg")img = cv2.cvtColor(frame, cv2.COLOR_RGB2GRAY)dets = detector(img, 1) print(dets) for index, det in enumerate(dets):    right_top = (det.right(), det.top())    left_bottom = (det.left(), det.bottom())    # 画面部检测框    # cv2.rectangle(frame, right_top, left_bottom, (255, 0, 0), 2)    print(right_top,left_bottom)     # 面部特征点检测    shape = predictor(img, det)    landmarks = np.matrix([[p.x, p.y] for p in shape.parts()])    for idx, point in enumerate(landmarks):        # 68点的坐标        pos = (point[0, 0], point[0, 1])        cv2.circle(frame, pos, 5, (0, 0, 255), -1)        cv2.putText(frame,str(idx),pos,cv2.FONT_HERSHEY_COMPLEX,0.6,(0,0,255),1) cv2.imwrite('kp2.jpg',frame)cv2.imshow('res',frame)cv2.waitKey()

2、dlib68点

基本包含脸部区域关键点，但是额头区域没有关键点以及只有眉毛的上 边缘点 。

import dlibimport numpy as npimport cv2import time predictor_path = '../model/shape_predictor_68_face_landmarks.dat'detector = dlib.get_frontal_face_detector()predictor = dlib.shape_predictor(predictor_path) start_time = time.time() frame = cv2.imread("../image/ab.jpg")img = cv2.cvtColor(frame, cv2.COLOR_RGB2GRAY)dets = detector(img, 1) print(dets) for index, det in enumerate(dets):    right_top = (det.right(), det.top())    left_bottom = (det.left(), det.bottom())    # 画面部检测框    # cv2.rectangle(frame, right_top, left_bottom, (255, 0, 0), 2)    print(right_top,left_bottom)     # 面部特征点检测    shape = predictor(img, det)    landmarks = np.matrix([[p.x, p.y] for p in shape.parts()])    for idx, point in enumerate(landmarks):        # 68点的坐标        pos = (point[0, 0], point[0, 1])        cv2.circle(frame, pos, 5, (0, 0, 255), -1)        cv2.putText(frame,str(idx),pos,cv2.FONT_HERSHEY_COMPLEX,0.6,(0,0,255),1) cv2.imwrite('kp2.jpg',frame)cv2.imshow('res',frame)cv2.waitKey()

3、 dlib81点

相比较68点，多出13点分布在额头区域，但是点的精度不是太高。

import dlibimport numpy as npimport cv2import time predictor_path = '../model/shape_predictor_81_face_landmarks.dat'detector = dlib.get_frontal_face_detector()predictor = dlib.shape_predictor(predictor_path) start_time = time.time() frame = cv2.imread("../image/ab.jpg")img = cv2.cvtColor(frame, cv2.COLOR_RGB2GRAY)dets = detector(img, 1) print(dets) for index, det in enumerate(dets):    right_top = (det.right(), det.top())    left_bottom = (det.left(), det.bottom())    # 画面部检测框    # cv2.rectangle(frame, right_top, left_bottom, (255, 0, 0), 2)    print(right_top,left_bottom)     # 面部特征点检测    shape = predictor(img, det)    landmarks = np.matrix([[p.x, p.y] for p in shape.parts()])    for idx, point in enumerate(landmarks):        # 68点的坐标        pos = (point[0, 0], point[0, 1])        cv2.circle(frame, pos, 5, (0, 0, 255), -1)        cv2.putText(frame,str(idx),pos,cv2.FONT_HERSHEY_COMPLEX,0.6,(0,0,255),1) cv2.imwrite('kp2.jpg',frame)cv2.imshow('res',frame)cv2.waitKey()

免责声明：本文系网络转载或改编，未找到原创作者，版权归原作者所有。如涉及版权，请联系删