483 lines
18 KiB
Python
Executable File
483 lines
18 KiB
Python
Executable File
# Copyright (C) 2018-2021 coneypo
|
|
# SPDX-License-Identifier: MIT
|
|
|
|
import os
|
|
import dlib
|
|
import csv
|
|
import numpy as np
|
|
import logging
|
|
import cv2
|
|
from PIL import Image
|
|
|
|
# 要读取人脸图像文件的路径
|
|
path_images_from_camera = "data/data_faces"
|
|
|
|
# Dlib 检测器
|
|
predictor = dlib.shape_predictor('models/dlib/shape_predictor_68_face_landmarks.dat')
|
|
face_reco_model = dlib.face_recognition_model_v1("models/dlib/dlib_face_recognition_resnet_model_v1.dat")
|
|
|
|
|
|
class DNNFaceDetector:
|
|
def __init__(self, confidence_threshold=0.7):
|
|
"""初始化OpenCV DNN人脸检测器"""
|
|
self.confidence_threshold = confidence_threshold
|
|
self.net = self.load_face_detection_model()
|
|
|
|
def load_face_detection_model(self):
|
|
"""加载OpenCV DNN人脸检测模型"""
|
|
try:
|
|
# 方法1: 尝试加载TensorFlow模型
|
|
model_path = "models/opencv_face_detector_uint8.pb"
|
|
config_path = "models/opencv_face_detector.pbtxt"
|
|
|
|
if os.path.exists(model_path) and os.path.exists(config_path):
|
|
net = cv2.dnn.readNetFromTensorflow(model_path, config_path)
|
|
print("使用 TensorFlow 人脸检测模型")
|
|
self.use_tensorflow = True
|
|
return net
|
|
except Exception as e:
|
|
print(f"TensorFlow模型加载失败: {e}")
|
|
|
|
try:
|
|
# 方法2: 尝试加载Caffe模型
|
|
proto_path = "models/deploy.prototxt"
|
|
model_path = "models/res10_300x300_ssd_iter_140000_fp16.caffemodel"
|
|
|
|
if os.path.exists(proto_path) and os.path.exists(model_path):
|
|
net = cv2.dnn.readNetFromCaffe(proto_path, model_path)
|
|
print("使用 Caffe 人脸检测模型")
|
|
self.use_tensorflow = False
|
|
return net
|
|
except Exception as e:
|
|
print(f"Caffe模型加载失败: {e}")
|
|
|
|
# 方法3: 使用OpenCV内置的Haar级联作为备选
|
|
print("使用 Haar Cascade 作为备选检测器")
|
|
self.use_haar = True
|
|
cascade_path ='models/haarcascade_frontalface_default.xml'
|
|
if os.path.exists(cascade_path):
|
|
return cv2.CascadeClassifier(cascade_path)
|
|
else:
|
|
# 如果内置路径不存在，尝试其他路径
|
|
possible_paths = [
|
|
'/usr/share/opencv4/haarcascades/haarcascade_frontalface_default.xml',
|
|
'/usr/share/opencv/haarcascades/haarcascade_frontalface_default.xml',
|
|
'haarcascade_frontalface_default.xml'
|
|
]
|
|
for path in possible_paths:
|
|
if os.path.exists(path):
|
|
return cv2.CascadeClassifier(path)
|
|
|
|
print("错误: 未找到任何人脸检测模型")
|
|
return None
|
|
|
|
def detect_faces(self, image):
|
|
"""使用DNN检测人脸"""
|
|
if hasattr(self, 'use_haar') and self.use_haar:
|
|
return self.detect_faces_haar(image)
|
|
|
|
if self.net is None:
|
|
return []
|
|
|
|
h, w = image.shape[:2]
|
|
|
|
# 创建blob输入
|
|
if hasattr(self, 'use_tensorflow') and self.use_tensorflow:
|
|
blob = cv2.dnn.blobFromImage(image, 1.0, (300, 300), [104, 117, 123], False, False)
|
|
else:
|
|
# Caffe模型
|
|
blob = cv2.dnn.blobFromImage(cv2.resize(image, (300, 300)), 1.0,
|
|
(300, 300), (104.0, 177.0, 123.0))
|
|
|
|
self.net.setInput(blob)
|
|
detections = self.net.forward()
|
|
|
|
faces = []
|
|
for i in range(detections.shape[2]):
|
|
confidence = detections[0, 0, i, 2]
|
|
|
|
if confidence > self.confidence_threshold:
|
|
# 提取边界框坐标
|
|
if hasattr(self, 'use_tensorflow') and self.use_tensorflow:
|
|
# TensorFlow模型输出格式
|
|
x1 = int(detections[0, 0, i, 3] * w)
|
|
y1 = int(detections[0, 0, i, 4] * h)
|
|
x2 = int(detections[0, 0, i, 5] * w)
|
|
y2 = int(detections[0, 0, i, 6] * h)
|
|
else:
|
|
# Caffe模型输出格式
|
|
x1 = int(detections[0, 0, i, 3] * w)
|
|
y1 = int(detections[0, 0, i, 4] * h)
|
|
x2 = int(detections[0, 0, i, 5] * w)
|
|
y2 = int(detections[0, 0, i, 6] * h)
|
|
|
|
# 确保坐标在图像范围内
|
|
x1, y1 = max(0, x1), max(0, y1)
|
|
x2, y2 = min(w, x2), min(h, y2)
|
|
|
|
# 转换为dlib矩形格式（保持兼容性）
|
|
if x2 > x1 and y2 > y1: # 确保是有效的矩形
|
|
face_rect = dlib.rectangle(x1, y1, x2, y2)
|
|
faces.append((face_rect, confidence))
|
|
|
|
return faces
|
|
|
|
def detect_faces_haar(self, image):
|
|
"""备选：使用Haar级联检测"""
|
|
gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
|
|
faces_cv = self.net.detectMultiScale(
|
|
gray,
|
|
scaleFactor=1.1,
|
|
minNeighbors=5,
|
|
minSize=(30, 30)
|
|
)
|
|
|
|
faces = []
|
|
for (x, y, w, h) in faces_cv:
|
|
face_rect = dlib.rectangle(x, y, x + w, y + h)
|
|
faces.append((face_rect, 0.9)) # Haar没有置信度
|
|
|
|
return faces
|
|
|
|
|
|
# 全局DNN检测器实例
|
|
dnn_detector = DNNFaceDetector(confidence_threshold=0.7)
|
|
|
|
class DNNFaceDetector:
|
|
def __init__(self, confidence_threshold=0.7):
|
|
"""初始化OpenCV DNN人脸检测器"""
|
|
self.confidence_threshold = confidence_threshold
|
|
self.net = self.load_face_detection_model()
|
|
|
|
def load_face_detection_model(self):
|
|
"""加载OpenCV DNN人脸检测模型"""
|
|
try:
|
|
# 方法1: 尝试加载TensorFlow模型
|
|
model_path = "models/opencv_face_detector_uint8.pb"
|
|
config_path = "models/opencv_face_detector.pbtxt"
|
|
|
|
if os.path.exists(model_path) and os.path.exists(config_path):
|
|
net = cv2.dnn.readNetFromTensorflow(model_path, config_path)
|
|
print("使用 TensorFlow 人脸检测模型")
|
|
self.use_tensorflow = True
|
|
return net
|
|
except Exception as e:
|
|
print(f"TensorFlow模型加载失败: {e}")
|
|
|
|
try:
|
|
# 方法2: 尝试加载Caffe模型
|
|
proto_path = "models/deploy.prototxt"
|
|
model_path = "models/res10_300x300_ssd_iter_140000_fp16.caffemodel"
|
|
|
|
if os.path.exists(proto_path) and os.path.exists(model_path):
|
|
net = cv2.dnn.readNetFromCaffe(proto_path, model_path)
|
|
print("使用 Caffe 人脸检测模型")
|
|
self.use_tensorflow = False
|
|
return net
|
|
except Exception as e:
|
|
print(f"Caffe模型加载失败: {e}")
|
|
|
|
# 方法3: 使用OpenCV内置的Haar级联作为备选
|
|
print("使用 Haar Cascade 作为备选检测器")
|
|
self.use_haar = True
|
|
cascade_path ='models/haarcascade_frontalface_default.xml'
|
|
if os.path.exists(cascade_path):
|
|
return cv2.CascadeClassifier(cascade_path)
|
|
else:
|
|
# 如果内置路径不存在，尝试其他路径
|
|
possible_paths = [
|
|
'/usr/share/opencv4/haarcascades/haarcascade_frontalface_default.xml',
|
|
'/usr/share/opencv/haarcascades/haarcascade_frontalface_default.xml',
|
|
'models/haarcascade_frontalface_default.xml'
|
|
]
|
|
for path in possible_paths:
|
|
if os.path.exists(path):
|
|
return cv2.CascadeClassifier(path)
|
|
|
|
print("错误: 未找到任何人脸检测模型")
|
|
return None
|
|
|
|
def detect_faces(self, image):
|
|
"""使用DNN检测人脸"""
|
|
if hasattr(self, 'use_haar') and self.use_haar:
|
|
return self.detect_faces_haar(image)
|
|
|
|
if self.net is None:
|
|
return []
|
|
|
|
h, w = image.shape[:2]
|
|
|
|
# 创建blob输入
|
|
if hasattr(self, 'use_tensorflow') and self.use_tensorflow:
|
|
blob = cv2.dnn.blobFromImage(image, 1.0, (300, 300), [104, 117, 123], False, False)
|
|
else:
|
|
# Caffe模型
|
|
blob = cv2.dnn.blobFromImage(cv2.resize(image, (300, 300)), 1.0,
|
|
(300, 300), (104.0, 177.0, 123.0))
|
|
|
|
self.net.setInput(blob)
|
|
detections = self.net.forward()
|
|
|
|
faces = []
|
|
for i in range(detections.shape[2]):
|
|
confidence = detections[0, 0, i, 2]
|
|
|
|
if confidence > self.confidence_threshold:
|
|
# 提取边界框坐标
|
|
if hasattr(self, 'use_tensorflow') and self.use_tensorflow:
|
|
# TensorFlow模型输出格式
|
|
x1 = int(detections[0, 0, i, 3] * w)
|
|
y1 = int(detections[0, 0, i, 4] * h)
|
|
x2 = int(detections[0, 0, i, 5] * w)
|
|
y2 = int(detections[0, 0, i, 6] * h)
|
|
else:
|
|
# Caffe模型输出格式
|
|
x1 = int(detections[0, 0, i, 3] * w)
|
|
y1 = int(detections[0, 0, i, 4] * h)
|
|
x2 = int(detections[0, 0, i, 5] * w)
|
|
y2 = int(detections[0, 0, i, 6] * h)
|
|
|
|
# 确保坐标在图像范围内
|
|
x1, y1 = max(0, x1), max(0, y1)
|
|
x2, y2 = min(w, x2), min(h, y2)
|
|
|
|
# 转换为dlib矩形格式（保持兼容性）
|
|
if x2 > x1 and y2 > y1: # 确保是有效的矩形
|
|
face_rect = dlib.rectangle(x1, y1, x2, y2)
|
|
faces.append((face_rect, confidence))
|
|
|
|
return faces
|
|
|
|
def detect_faces_haar(self, image):
|
|
"""备选：使用Haar级联检测"""
|
|
gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
|
|
faces_cv = self.net.detectMultiScale(
|
|
gray,
|
|
scaleFactor=1.1,
|
|
minNeighbors=5,
|
|
minSize=(30, 30)
|
|
)
|
|
|
|
faces = []
|
|
for (x, y, w, h) in faces_cv:
|
|
face_rect = dlib.rectangle(x, y, x + w, y + h)
|
|
faces.append((face_rect, 0.9)) # Haar没有置信度
|
|
|
|
return faces
|
|
|
|
|
|
# 全局DNN检测器实例
|
|
dnn_detector = DNNFaceDetector(confidence_threshold=0.7)
|
|
|
|
def return_128d_features(path_img):
|
|
"""返回单张图像的 128D 特征（使用OpenCV DNN检测）"""
|
|
"""返回单张图像的 128D 特征（使用OpenCV DNN检测）"""
|
|
try:
|
|
# 直接使用OpenCV读取图像，避免PIL转换
|
|
img_rd = cv2.imread(path_img)
|
|
if img_rd is None:
|
|
logging.error("无法读取图像: %s", path_img)
|
|
return None
|
|
|
|
# 对于大图像，先缩小以提高检测速度
|
|
h, w = img_rd.shape[:2]
|
|
if w > 1000 or h > 1000:
|
|
scale = 1000 / max(w, h)
|
|
small_img = cv2.resize(img_rd, (int(w * scale), int(h * scale)))
|
|
face_results = dnn_detector.detect_faces(small_img)
|
|
|
|
# 缩放坐标回原图
|
|
scaled_faces = []
|
|
for face_rect, confidence in face_results:
|
|
scaled_face = dlib.rectangle(
|
|
int(face_rect.left() / scale),
|
|
int(face_rect.top() / scale),
|
|
int(face_rect.right() / scale),
|
|
int(face_rect.bottom() / scale)
|
|
)
|
|
scaled_faces.append((scaled_face, confidence))
|
|
face_results = scaled_faces
|
|
else:
|
|
face_results = dnn_detector.detect_faces(img_rd)
|
|
|
|
if len(face_results) != 0:
|
|
# 取置信度最高的人脸
|
|
best_face = max(face_results, key=lambda x: x[1])
|
|
face_rect = best_face[0]
|
|
|
|
logging.info("%-40s %-20s", "检测到人脸的图像:", path_img)
|
|
logging.info("检测置信度: %.3f", best_face[1])
|
|
|
|
# 使用dlib进行特征点检测和特征提取
|
|
shape = predictor(img_rd, face_rect)
|
|
# 直接使用OpenCV读取图像，避免PIL转换
|
|
img_rd = cv2.imread(path_img)
|
|
if img_rd is None:
|
|
logging.error("无法读取图像: %s", path_img)
|
|
return None
|
|
|
|
# 对于大图像，先缩小以提高检测速度
|
|
h, w = img_rd.shape[:2]
|
|
if w > 1000 or h > 1000:
|
|
scale = 1000 / max(w, h)
|
|
small_img = cv2.resize(img_rd, (int(w * scale), int(h * scale)))
|
|
face_results = dnn_detector.detect_faces(small_img)
|
|
|
|
# 缩放坐标回原图
|
|
scaled_faces = []
|
|
for face_rect, confidence in face_results:
|
|
scaled_face = dlib.rectangle(
|
|
int(face_rect.left() / scale),
|
|
int(face_rect.top() / scale),
|
|
int(face_rect.right() / scale),
|
|
int(face_rect.bottom() / scale)
|
|
)
|
|
scaled_faces.append((scaled_face, confidence))
|
|
face_results = scaled_faces
|
|
else:
|
|
face_results = dnn_detector.detect_faces(img_rd)
|
|
|
|
if len(face_results) != 0:
|
|
# 取置信度最高的人脸
|
|
best_face = max(face_results, key=lambda x: x[1])
|
|
face_rect = best_face[0]
|
|
|
|
logging.info("%-40s %-20s", "检测到人脸的图像:", path_img)
|
|
logging.info("检测置信度: %.3f", best_face[1])
|
|
|
|
# 使用dlib进行特征点检测和特征提取
|
|
shape = predictor(img_rd, face_rect)
|
|
face_descriptor = face_reco_model.compute_face_descriptor(img_rd, shape)
|
|
|
|
|
|
return face_descriptor
|
|
else:
|
|
logging.warning("未检测到人脸: %s", path_img)
|
|
return None
|
|
|
|
|
|
except Exception as e:
|
|
logging.error("处理图像时出错 %s: %s", path_img, e)
|
|
return None
|
|
|
|
def return_features_mean_personX(path_face_personX):
|
|
"""返回 personX 的 128D 特征均值"""
|
|
features_list_personX = []
|
|
photos_list = os.listdir(path_face_personX)
|
|
|
|
if photos_list:
|
|
for photo in photos_list:
|
|
# 只处理图像文件
|
|
if not photo.lower().endswith(('.jpg', '.jpeg', '.png', '.bmp')):
|
|
continue
|
|
|
|
# 只处理图像文件
|
|
if not photo.lower().endswith(('.jpg', '.jpeg', '.png', '.bmp')):
|
|
continue
|
|
|
|
photo_path = os.path.join(path_face_personX, photo)
|
|
logging.info("正在读取图像: %s", photo_path)
|
|
|
|
features_128d = return_128d_features(photo_path)
|
|
if features_128d is not None:
|
|
features_list_personX.append(features_128d)
|
|
else:
|
|
logging.warning("无法从图像中提取特征: %s", photo_path)
|
|
else:
|
|
logging.warning("文件夹为空: %s", path_face_personX)
|
|
|
|
# 计算 128D 特征的均值
|
|
if features_list_personX:
|
|
features_mean_personX = np.array(features_list_personX).mean(axis=0)
|
|
logging.info("成功提取 %d 张人脸特征", len(features_list_personX))
|
|
else:
|
|
features_mean_personX = np.zeros(128, dtype=np.float64)
|
|
logging.warning("未提取到任何特征: %s", path_face_personX)
|
|
|
|
return features_mean_personX
|
|
|
|
|
|
def get_person_name_from_folder(folder_name):
|
|
"""从文件夹名称获取有意义的姓名"""
|
|
# 常见的文件夹前缀
|
|
prefixes = ['person_', 'face_', 'user_']
|
|
|
|
for prefix in prefixes:
|
|
if folder_name.startswith(prefix):
|
|
name_part = folder_name[len(prefix):]
|
|
# 如果剩下的部分是纯数字，使用完整文件夹名
|
|
if name_part.isdigit():
|
|
return folder_name
|
|
else:
|
|
return name_part
|
|
|
|
return folder_name
|
|
|
|
def main():
|
|
logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')
|
|
# 检查源文件夹是否存在
|
|
if not os.path.exists(path_images_from_camera):
|
|
logging.error("人脸图像文件夹不存在: %s", path_images_from_camera)
|
|
return
|
|
|
|
# 获取人脸文件夹列表
|
|
person_list = os.listdir(path_images_from_camera)
|
|
person_list.sort()
|
|
|
|
if not person_list:
|
|
logging.error("没有人脸文件夹可处理")
|
|
return
|
|
|
|
logging.info("找到 %d 个人脸文件夹: %s", len(person_list), person_list)
|
|
|
|
# 创建CSV文件
|
|
with open("data/features_all.csv", "w", newline="", encoding="utf-8") as csvfile:
|
|
writer = csv.writer(csvfile)
|
|
|
|
successful_count = 0
|
|
for person_folder in person_list:
|
|
folder_path = os.path.join(path_images_from_camera, person_folder)
|
|
|
|
if not os.path.isdir(folder_path):
|
|
logging.warning("跳过非文件夹: %s", person_folder)
|
|
logging.warning("跳过非文件夹: %s", person_folder)
|
|
continue
|
|
|
|
logging.info("=" * 60)
|
|
logging.info("=" * 60)
|
|
logging.info("处理文件夹: %s", person_folder)
|
|
|
|
# 提取特征
|
|
features_mean = return_features_mean_personX(folder_path)
|
|
|
|
# 获取有意义的姓名
|
|
person_name = get_person_name_from_folder(person_folder)
|
|
logging.info("使用姓名: %s", person_name)
|
|
|
|
# 检查特征是否有效（非全零）
|
|
if np.all(features_mean == 0):
|
|
logging.warning("特征提取失败，跳过: %s", person_folder)
|
|
continue
|
|
|
|
# 检查特征是否有效（非全零）
|
|
if np.all(features_mean == 0):
|
|
logging.warning("特征提取失败，跳过: %s", person_folder)
|
|
continue
|
|
|
|
# 构建行数据：姓名 + 128维特征
|
|
row_data = [person_name] + features_mean.tolist()
|
|
writer.writerow(row_data)
|
|
|
|
successful_count += 1
|
|
logging.info("完成: %s", person_name)
|
|
logging.info("-" * 50)
|
|
|
|
logging.info("=" * 60)
|
|
logging.info("处理完成: 成功 %d/%d 个人脸文件夹", successful_count, len(person_list))
|
|
logging.info("=" * 60)
|
|
logging.info("处理完成: 成功 %d/%d 个人脸文件夹", successful_count, len(person_list))
|
|
logging.info("特征数据已保存到: data/features_all.csv")
|
|
|
|
|
|
if __name__ == '__main__':
|
|
main() |