人脸姿态估计(代码已跑通)

63 阅读 0 评论 42 点赞

我是靠谱客的博主温柔香烟，最近开发中收集的这篇文章主要介绍人脸姿态估计(代码已跑通)，觉得挺不错的，现在分享给大家，希望可以做个参考。

概述

人脸姿态估计主要是获得脸部朝向的角度信息。一般可以用旋转矩阵、旋转向量、四元数或欧拉角表示（这四个量也可以互相转换）。一般而言，欧拉角可读性更好一些，使用更为广泛。本文获得的人脸姿态信息用三个欧拉角（pitch，yaw，roll）表示,通俗讲就是抬头、摇头和转头。

话不多说，先上图：

代码部分解答：

本例中只是用了14个关键点，主要是因为通用的3D人脸模型的标注方式和dlib检测出来的点不对应，因此选择了部分共同的特征，points_68 代表的部分检测出来的点，排列方式与model_points 对应。

代码如下：

import cv2
import matplotlib.pyplot as plt
import numpy as np
import dlib
import os
import sys
import math
model_points = np.array([
            [6.825897, 6.760612, 4.402142],
            [1.330353, 7.122144, 6.903745],
            [-1.330353, 7.122144, 6.903745 ],
            [-6.825897, 6.760612, 4.402142],
            [5.311432, 5.485328, 3.987654],
            [1.789930, 5.393625, 4.413414],
            [-1.789930, 5.393625, 4.413414],
            [-5.311432, 5.485328, 3.987654],
            [2.005628, 1.409845, 6.165652],
            [-2.005628, 1.409845, 6.165652],
            [2.774015, -2.080775, 5.048531],
            [-2.774015, -2.080775, 5.048531],
            [0.000000, -3.116408, 6.097667],
            [0.000000, -7.415691, 4.070434]])
detector = dlib.get_frontal_face_detector() 
pic_path = "/media/xxxxx/5874eb32-3573-4145-9cce-b105957df272/data_process/data_pro/solov2/5007442_2.jpg"
img = cv2.imread(pic_path) 
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
faces = detector(img, 2) 
face = faces[0] 
x0,y0,x1,y1 = face.left(),face.top(),face.right(),face.bottom()
cv2.rectangle(img, (x0,y0), (x1,y1), (255,0,0), 2) #画个人脸框框
predictor = dlib.shape_predictor('/media/zhoukx/5874eb32-3573-4145-9cce-b105957df272/data_process/data_pro/solov2/shape_predictor_68_face_landmarks.dat') 
ldmk = predictor(img, face) 


points = np.array([(p.x, p.y) for p in ldmk.parts()],dtype="double")
#17 left brow left corner
#21 left brow right corner
#22 right brow left corner
#26 right brow right corner
#36 left eye left corner
#39 left eye right corner
#42 right eye left corner
#45 right eye right corner
#31 nose left corner
#35 nose right corner
#48 mouth left corner
#54 mouth right corner
#57 mouth central bottom corner
#8 chin corner

points_68 = np.array([
        points[17],
        points[21],
        points[22],
        points[26],
        points[36],
        points[39],
        points[42],
        points[45],
        points[31],
        points[35],
        points[48],
        points[54],
        points[57],
        points[8]
],dtype="double")
size=img.shape
focal_length = size[1]
center = (size[1]/2, size[0]/2)
camera_matrix = np.array(
                         [[focal_length, 0, center[0]],
                         [0, focal_length, center[1]],
                         [0, 0, 1]], dtype = "double"
                         )
 
print ("Camera Matrix :n {0}".format(camera_matrix))

dist_coeffs = np.zeros((4,1)) # Assuming no lens distortion
(success, rotation_vector, translation_vector) = cv2.solvePnP(model_points, points_68, camera_matrix, dist_coeffs)
 
print ("Rotation Vector:n {0}".format(rotation_vector))
print ("Translation Vector:n {0}".format(translation_vector))

theta = np.linalg.norm(rotation_vector)
r = rotation_vector / theta
R_ = np.array([[0, -r[2][0], r[1][0]],
               [r[2][0], 0, -r[0][0]],
               [-r[1][0], r[0][0], 0]])
R = np.cos(theta) * np.eye(3) + (1 - np.cos(theta)) * r * r.T + np.sin(theta) * R_
print('旋转矩阵')
print(R)
def isRotationMatrix(R):
    Rt = np.transpose(R)  
    shouldBeIdentity = np.dot(Rt, R)   
    I = np.identity(3, dtype=R.dtype)         
    n = np.linalg.norm(I - shouldBeIdentity)   
    return n < 1e-6                           
 
 
def rotationMatrixToAngles(R):
    assert (isRotationMatrix(R))   
 
    sy = math.sqrt(R[0, 0] * R[0, 0] + R[1, 0] * R[1, 0])  
 
    singular = sy < 1e-6  
 
    if not singular:  
        x = math.atan2(R[2, 1], R[2, 2])
        y = math.atan2(-R[2, 0], sy)
        z = math.atan2(R[1, 0], R[0, 0])
    else:            
        x = math.atan2(-R[1, 2], R[1, 1])
        y = math.atan2(-R[2, 0], sy)  
        z = 0
    
    x = x*180.0/3.141592653589793
    y = y*180.0/3.141592653589793
    z = z*180.0/3.141592653589793

   
    line = 'pitch:{:.1f}nyaw:{:.1f}nroll:{:.1f}'.format(y,x,z)
    print('{},{}'.format(os.path.basename(pic_path), line.replace('n',',')))
 
    y = 20
    for _, txt in enumerate(line.split('n')):
        cv2.putText(img, txt, (20, y), cv2.FONT_HERSHEY_PLAIN, 1.3, (0,0,255), 1)
        y = y + 15
 
    for p in points_68:
        cv2.circle(img, (int(p[0]),int(p[1])), 2, (0,255,0), -1, 0)
 
    cv2.imshow('img', img)
    if cv2.waitKey(-1) == 27:
        pass
 
    return np.array([x, y, z])
n = isRotationMatrix(R)
x_y_z = rotationMatrixToAngles(R)
print(x_y_z)

参考文章：

1 人脸姿态估计（计算欧拉角）_sunnygirl's house-CSDN博客_人脸姿态估计

2 关于人脸关键点的数据集WFLW数据预处理_baidu_40840693的博客-CSDN博客

3 GitHub - lincolnhard/head-pose-estimation: Real-time head pose estimation built with OpenCV and dlib

4 重磅！头部姿态估计「原理详解 + 实战代码」来啦！