225 lines
7.6 KiB
Python
225 lines
7.6 KiB
Python
import os
|
||
import cv2
|
||
import numpy as np
|
||
import face_recognition
|
||
from pathlib import Path
|
||
import time
|
||
|
||
class FaceDetector:
|
||
def __init__(self, conf_thres=0.5):
|
||
"""初始化人脸检测器，使用face_recognition库
|
||
|
||
Args:
|
||
conf_thres: 置信度阈值
|
||
"""
|
||
self.conf_thres = conf_thres
|
||
|
||
def _load_model(self):
|
||
"""不再需要加载模型，使用face_recognition库"""
|
||
print("使用face_recognition库进行人脸检测，无需加载模型")
|
||
pass
|
||
|
||
def detect_faces(self, image):
|
||
"""使用face_recognition库检测图像中的人脸
|
||
|
||
Args:
|
||
image: 输入图像 (numpy数组，BGR格式)
|
||
|
||
Returns:
|
||
list: 检测到的人脸边界框列表 [x1, y1, x2, y2, confidence]
|
||
"""
|
||
boxes = []
|
||
|
||
# 使用face_recognition库检测人脸位置
|
||
# 将BGR图像转换为RGB格式（face_recognition需要RGB格式）
|
||
rgb_image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
|
||
|
||
# 使用face_recognition检测人脸位置
|
||
face_locations = face_recognition.face_locations(rgb_image)
|
||
|
||
# 将检测结果转换为所需的格式 [x1, y1, x2, y2, confidence]
|
||
for face_location in face_locations:
|
||
# face_location格式为(top, right, bottom, left)
|
||
top, right, bottom, left = face_location
|
||
|
||
# 转换为[x1, y1, x2, y2, confidence]格式
|
||
# 由于face_recognition不提供置信度，我们使用1.0作为默认值
|
||
boxes.append([left, top, right, bottom, 1.0])
|
||
|
||
print(f"检测到 {len(boxes)} 个人脸")
|
||
|
||
# 如果未检测到人脸，尝试使用OpenCV的人脸检测器作为备份
|
||
if len(boxes) == 0:
|
||
try:
|
||
print("使用OpenCV备用人脸检测...")
|
||
face_cascade = cv2.CascadeClassifier(cv2.data.haarcascades + 'haarcascade_frontalface_default.xml')
|
||
gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
|
||
faces = face_cascade.detectMultiScale(gray, 1.1, 5)
|
||
|
||
for (x, y, w, h) in faces:
|
||
boxes.append([x, y, x+w, y+h, 0.8]) # 使用默认置信度0.8
|
||
print(f"OpenCV检测到 {len(faces)} 个人脸")
|
||
except Exception as e:
|
||
print(f"OpenCV备用人脸检测失败: {e}")
|
||
|
||
return boxes
|
||
|
||
class FaceRecognizer:
|
||
def __init__(self, detector=None, similarity_threshold=0.6):
|
||
"""初始化人脸识别器
|
||
|
||
Args:
|
||
detector: 人脸检测器实例，如果为None则创建默认检测器
|
||
similarity_threshold: 人脸相似度阈值
|
||
"""
|
||
self.detector = detector if detector else FaceDetector()
|
||
self.similarity_threshold = similarity_threshold
|
||
# 使用OpenCV内置的人脸识别器
|
||
self.face_recognizer = cv2.face.LBPHFaceRecognizer_create()
|
||
self.face_size = (128, 128) # 标准化人脸大小
|
||
|
||
def extract_face_features(self, image, box=None):
|
||
"""提取人脸特征
|
||
|
||
Args:
|
||
image: 输入图像
|
||
box: 可选的人脸边界框 [x1, y1, x2, y2]，如果未提供则自动检测
|
||
|
||
Returns:
|
||
numpy.ndarray: 人脸特征向量
|
||
"""
|
||
# 如果没有提供边界框，使用检测器检测人脸
|
||
if box is None:
|
||
boxes = self.detector.detect_faces(image)
|
||
if not boxes:
|
||
return None
|
||
box = boxes[0][:4] # 使用第一个检测到的人脸
|
||
|
||
# 裁剪人脸区域
|
||
x1, y1, x2, y2 = box
|
||
face_img = image[y1:y2, x1:x2]
|
||
|
||
# 转换为灰度图像
|
||
face_gray = cv2.cvtColor(face_img, cv2.COLOR_BGR2GRAY)
|
||
|
||
# 调整大小为标准尺寸
|
||
face_resized = cv2.resize(face_gray, self.face_size)
|
||
|
||
# 使用直方图均衡化增强对比度
|
||
face_normalized = cv2.equalizeHist(face_resized)
|
||
|
||
# 简单特征提取：将图像展平为向量
|
||
# 在实际应用中，可以使用更复杂的特征提取方法，如HOG或深度特征
|
||
feature_vector = face_normalized.flatten().astype(np.float32)
|
||
|
||
# 对特征向量进行标准化
|
||
if np.linalg.norm(feature_vector) > 0:
|
||
feature_vector = feature_vector / np.linalg.norm(feature_vector)
|
||
|
||
return feature_vector
|
||
|
||
def compare_faces(self, known_feature, unknown_feature):
|
||
"""比较两个人脸特征的相似度
|
||
|
||
Args:
|
||
known_feature: 已知人脸特征
|
||
unknown_feature: 待比对人脸特征
|
||
|
||
Returns:
|
||
float: 相似度 (0-1之间，越大表示越相似)
|
||
"""
|
||
if known_feature is None or unknown_feature is None:
|
||
return 0.0
|
||
|
||
# 计算欧氏距离
|
||
face_distance = np.linalg.norm(known_feature - unknown_feature)
|
||
# 将距离转换为相似度
|
||
similarity = 1.0 / (1.0 + face_distance)
|
||
|
||
return similarity
|
||
|
||
def identify_face(self, image, known_faces, box=None):
|
||
"""识别人脸
|
||
|
||
Args:
|
||
image: 输入图像
|
||
known_faces: 已知人脸特征列表 [(person_id, name, feature_vector), ...]
|
||
box: 可选的人脸边界框
|
||
|
||
Returns:
|
||
tuple or None: (person_id, name, similarity) 或 None（如果未匹配）
|
||
"""
|
||
# 提取人脸特征
|
||
face_feature = self.extract_face_features(image, box)
|
||
if face_feature is None:
|
||
return None
|
||
|
||
# 寻找最佳匹配
|
||
best_match = None
|
||
best_similarity = 0
|
||
|
||
for person_id, name, known_feature in known_faces:
|
||
similarity = self.compare_faces(known_feature, face_feature)
|
||
|
||
if similarity > best_similarity:
|
||
best_similarity = similarity
|
||
best_match = (person_id, name, similarity)
|
||
|
||
# 如果最佳匹配的相似度低于阈值，返回None
|
||
if best_match and best_match[2] >= self.similarity_threshold:
|
||
return best_match
|
||
|
||
return None
|
||
|
||
def draw_face_box(image, box, name=None, similarity=None, color=(0, 255, 0), thickness=2):
|
||
"""在图像上绘制人脸边界框和标签
|
||
|
||
Args:
|
||
image: 输入图像
|
||
box: 人脸边界框 [x1, y1, x2, y2, ...]
|
||
name: 可选的名称标签
|
||
similarity: 可选的相似度标签
|
||
color: 边界框颜色 (BGR格式)
|
||
thickness: 边界框线条粗细
|
||
|
||
Returns:
|
||
numpy.ndarray: 绘制了边界框的图像
|
||
"""
|
||
x1, y1, x2, y2 = box[:4]
|
||
|
||
# 绘制边界框
|
||
cv2.rectangle(image, (x1, y1), (x2, y2), color, thickness)
|
||
|
||
# 如果提供了名称，则绘制标签
|
||
if name or similarity is not None:
|
||
label = ""
|
||
if name:
|
||
label += name
|
||
if similarity is not None:
|
||
label += f" ({similarity:.2f})"
|
||
|
||
# 设置标签背景
|
||
(label_width, label_height), baseline = cv2.getTextSize(
|
||
label, cv2.FONT_HERSHEY_SIMPLEX, 0.5, 1
|
||
)
|
||
cv2.rectangle(
|
||
image,
|
||
(x1, y1 - label_height - 5),
|
||
(x1 + label_width, y1),
|
||
color,
|
||
-1 # 填充矩形
|
||
)
|
||
|
||
# 绘制标签文本
|
||
cv2.putText(
|
||
image,
|
||
label,
|
||
(x1, y1 - 5),
|
||
cv2.FONT_HERSHEY_SIMPLEX,
|
||
0.5,
|
||
(0, 0, 0), # 黑色文本
|
||
1
|
||
)
|
||
|
||
return image
|