feat(face): 实现基于InsightFace的人脸识别系统

- 集成InsightFace模型用于人脸检测和特征提取
- 实现人脸特征向量的余弦相似度计算- 构建SQLite数据库存储和管理人脸特征数据- 支持单个人脸注册、批量注册和人脸识别功能
- 添加人脸数据的增删查功能- 实现从文件夹自动批量注册人脸
- 支持设置识别阈值以提高识别准确性
- 添加详细的日志输出和错误处理机制
This commit is contained in:
mshe 2025-09-28 01:41:31 +08:00
parent c5c92f5245
commit dc8eed5ec7
9 changed files with 369 additions and 2 deletions

View File

@ -8,7 +8,7 @@ import numpy as np
# 加载训练好的模型 # 加载训练好的模型
# best.pt: 最佳模型,适用于生产 # best.pt: 最佳模型,适用于生产
# last.pt: 最后一轮训练的模型,适用于继续训练 # last.pt: 最后一轮训练的模型,适用于继续训练
yolo = YOLO('runs/detect/train7/weights/best.pt') yolo = YOLO('models/face/best.pt')
# 指定屏幕范围 # 指定屏幕范围
# x,y,width,height 全屏None # x,y,width,height 全屏None

42
16.人脸比对.py Normal file
View File

@ -0,0 +1,42 @@
import insightface
from insightface.app import FaceAnalysis
import cv2
# 1. 创建人脸分析应用
# 这会自动下载预训练模型(第一次运行时会下载,约 200MB
app = FaceAnalysis(name='buffalo_l') # 'buffalo_l' 是当前最准的模型集合
app.prepare(ctx_id=0, det_size=(640, 640)) # ctx_id=0 表示使用CPU ctx_id=1/gpu_id 表示使用GPU
# 2. 加载图片
img1 = cv2.imread('./resources/face/01.png')
# img2 = cv2.imread('./resources/face/02.png')
img2 = cv2.imread('./resources/face/16.jpg')
# 3. 进行人脸检测和特征提取
faces1 = app.get(img1)
faces2 = app.get(img2)
# 检查是否检测到人脸
if len(faces1) == 0 or len(faces2) == 0:
print("未在图片中检测到人脸!")
exit()
# 取每张图片中检测到的第一个人脸
face1 = faces1[0]
face2 = faces2[0]
# 4. 获取人脸特征向量(嵌入)
embedding1 = face1.embedding
embedding2 = face2.embedding
# 5. 计算特征向量之间的相似度(使用余弦相似度)
from numpy.linalg import norm
cos_sim = embedding1 @ embedding2.T / (norm(embedding1) * norm(embedding2))
print(f"人脸相似度(余弦): {cos_sim:.4f}")
# 6. 根据阈值判断是否为同一个人
threshold = 0.6 # 常用阈值可根据场景调整范围一般在0.2-0.8之间,值越大判断越严格)
if cos_sim > threshold:
print("判断为同一个人!")
else:
print("判断为不同人。")

59
17.人脸识别.py Normal file
View File

@ -0,0 +1,59 @@
import insightface
from insightface.app import FaceAnalysis
import cv2
import numpy as np
from numpy.linalg import norm
# 初始化模型
app = FaceAnalysis(name='buffalo_l')
app.prepare(ctx_id=0, det_size=(640, 640))
# 创建一个数据库来存储已知人脸的特征和姓名
face_database = {}
# --- 注册阶段:将已知人脸加入数据库 ---
def register_face(image_path, person_name):
img = cv2.imread(image_path)
faces = app.get(img)
if len(faces) != 1:
print(f"警告:在 {image_path} 中未检测到或检测到多张人脸,跳过。")
return
embedding = faces[0].embedding
face_database[person_name] = embedding
print(f"成功注册:{person_name}")
# 注册多个人
img1 = cv2.imread('./resources/face/01.png')
register_face('./resources/face/01.png', '01')
register_face('./resources/face/02.png', '02')
# --- 识别阶段:识别未知图片中的人 ---
def recognize_face(image_path):
img = cv2.imread(image_path)
faces = app.get(img)
if len(faces) == 0:
print("未检测到人脸。")
return
for i, face in enumerate(faces):
unknown_embedding = face.embedding
max_sim = -1
identity = "未知"
# 与数据库中的每个人脸进行比对
for name, known_embedding in face_database.items():
cos_sim = unknown_embedding @ known_embedding.T / (norm(unknown_embedding) * norm(known_embedding))
if cos_sim > max_sim:
max_sim = cos_sim
identity = name
# 根据阈值决定最终身份
threshold = 0.6
if max_sim < threshold:
identity = "未知"
print(f"人脸 {i+1}: 识别为 【{identity}】, 相似度:{max_sim:.4f}")
# 识别一张新图片
recognize_face('./resources/face/16.jpg')

View File

@ -0,0 +1,265 @@
import insightface
from insightface.app import FaceAnalysis
import cv2
import numpy as np
import sqlite3
import os
from numpy.linalg import norm
# 初始化InsightFace模型
app = FaceAnalysis(name='buffalo_l')
app.prepare(ctx_id=0, det_size=(640, 640))
class FaceDatabase:
def __init__(self, db_path='face_database.db'):
self.db_path = db_path
self.init_database()
def init_database(self):
"""初始化数据库"""
conn = sqlite3.connect(self.db_path)
cursor = conn.cursor()
cursor.execute('''
CREATE TABLE IF NOT EXISTS faces (
id INTEGER PRIMARY KEY AUTOINCREMENT,
name TEXT UNIQUE NOT NULL,
embedding BLOB NOT NULL,
created_time TIMESTAMP DEFAULT CURRENT_TIMESTAMP
)
''')
conn.commit()
conn.close()
print(f"数据库初始化完成: {self.db_path}")
def add_face(self, name, embedding):
"""添加人脸到数据库"""
conn = sqlite3.connect(self.db_path)
cursor = conn.cursor()
# 将numpy数组转换为bytes
embedding_blob = embedding.tobytes()
try:
cursor.execute(
'INSERT INTO faces (name, embedding) VALUES (?, ?)',
(name, embedding_blob)
)
conn.commit()
print(f"✅ 成功注册: {name}")
return True
except sqlite3.IntegrityError:
print(f"⚠️ 姓名已存在: {name}")
return False
finally:
conn.close()
def get_all_faces(self):
"""获取所有人脸数据"""
conn = sqlite3.connect(self.db_path)
cursor = conn.cursor()
cursor.execute('SELECT name, embedding FROM faces')
results = cursor.fetchall()
conn.close()
face_database = {}
for name, embedding_blob in results:
# 将bytes转换回numpy数组
embedding = np.frombuffer(embedding_blob, dtype=np.float32)
face_database[name] = embedding
print(f"📊 从数据库加载了 {len(face_database)} 个人脸特征")
return face_database
def delete_face(self, name):
"""删除人脸"""
conn = sqlite3.connect(self.db_path)
cursor = conn.cursor()
cursor.execute('DELETE FROM faces WHERE name = ?', (name,))
affected_rows = cursor.rowcount
conn.commit()
conn.close()
if affected_rows > 0:
print(f"🗑️ 已删除: {name}")
return True
else:
print(f"❌ 未找到: {name}")
return False
def face_exists(self, name):
"""检查人脸是否存在"""
conn = sqlite3.connect(self.db_path)
cursor = conn.cursor()
cursor.execute('SELECT 1 FROM faces WHERE name = ?', (name,))
exists = cursor.fetchone() is not None
conn.close()
return exists
# 创建全局数据库实例
face_db = FaceDatabase()
def register_face(image_path, person_name):
"""注册人脸"""
# 检查图片是否存在
if not os.path.exists(image_path):
print(f"❌ 图片不存在: {image_path}")
return False
# 读取图片
img = cv2.imread(image_path)
if img is None:
print(f"❌ 无法读取图片: {image_path}")
return False
# 使用InsightFace检测人脸
faces = app.get(img)
if len(faces) == 0:
print(f"❌ 在 {image_path} 中未检测到人脸")
return False
elif len(faces) > 1:
print(f"⚠️ 在 {image_path} 中检测到 {len(faces)} 张人脸,使用第一张")
# 提取人脸特征
embedding = faces[0].embedding
print(f"📐 提取到特征向量,维度: {embedding.shape}")
# 保存到数据库
return face_db.add_face(person_name, embedding)
def register_faces_from_folder(folder_path):
"""从文件夹批量注册人脸"""
if not os.path.exists(folder_path):
print(f"❌ 文件夹不存在: {folder_path}")
return
supported_formats = ('.jpg', '.jpeg', '.png', '.bmp', '.tiff')
registered_count = 0
for filename in os.listdir(folder_path):
if filename.lower().endswith(supported_formats):
# 使用文件名(不含扩展名)作为人名
person_name = os.path.splitext(filename)[0]
image_path = os.path.join(folder_path, filename)
if register_face(image_path, person_name):
registered_count += 1
print(f"🎉 批量注册完成,共注册 {registered_count} 个人脸")
def recognize_face(image_path, threshold=0.6):
"""识别人脸"""
# 检查图片是否存在
if not os.path.exists(image_path):
print(f"❌ 图片不存在: {image_path}")
return
# 读取图片
img = cv2.imread(image_path)
if img is None:
print(f"❌ 无法读取图片: {image_path}")
return
# 从数据库加载所有人脸特征
face_database = face_db.get_all_faces()
if not face_database:
print("❌ 数据库中无人脸数据,请先注册人脸")
return
# 使用InsightFace检测人脸
faces = app.get(img)
if len(faces) == 0:
print("❌ 未检测到人脸")
return
print(f"🔍 检测到 {len(faces)} 张人脸,开始识别...")
for i, face in enumerate(faces):
unknown_embedding = face.embedding
max_sim = -1
identity = "未知"
best_match_name = None
# 与数据库中的每个人脸进行比对
for name, known_embedding in face_database.items():
# 计算余弦相似度
cos_sim = unknown_embedding @ known_embedding.T / (norm(unknown_embedding) * norm(known_embedding))
if cos_sim > max_sim:
max_sim = cos_sim
identity = name
best_match_name = name
# 根据阈值决定最终身份
if max_sim < threshold:
identity = "未知"
# 输出结果
status = "" if identity != "未知" else ""
print(f"{status} 人脸 {i + 1}: 识别为 【{identity}】, 相似度: {max_sim:.4f}")
# 显示匹配详情
if identity != "未知":
matched_embedding = face_database[best_match_name]
print(f" 匹配特征: {best_match_name}, 范数: {norm(matched_embedding):.4f}")
def list_registered_faces():
"""列出所有已注册的人脸"""
face_database = face_db.get_all_faces()
if not face_database:
print("📭 数据库为空")
return
print(f"\n📋 已注册的人脸 ({len(face_database)} 个):")
for i, name in enumerate(face_database.keys(), 1):
print(f" {i}. {name}")
def delete_face(person_name):
"""删除指定人脸"""
return face_db.delete_face(person_name)
# 使用示例
if __name__ == "__main__":
print("=" * 50)
print("🎭 InsightFace 人脸识别系统 (SQLite版本)")
print("=" * 50)
# 1. 批量注册人脸
print("\n1. 📁 批量注册人脸")
register_faces_from_folder('./resources/face')
# 2. 单个注册
print("\n2. 👤 单个注册")
register_face('./resources/face/01.png', '张三')
register_face('./resources/face/02.png', '李四')
# 3. 查看已注册的人脸
print("\n3. 📊 已注册人脸列表")
list_registered_faces()
# 4. 识别人脸
print("\n4. 🔍 人脸识别测试")
recognize_face('./resources/face/16.jpg', threshold=0.6)
# 5. 删除测试
# print("\n5. 🗑️ 删除人脸测试")
# delete_face('测试删除')
print("\n" + "=" * 50)
print("✨ 程序执行完成")
print("=" * 50)

BIN
face_database.db Normal file

Binary file not shown.

View File

@ -4,4 +4,5 @@ numpy~=2.0.2
pyautogui~=0.9.54 pyautogui~=0.9.54
moviepy~=2.2.1 moviepy~=2.2.1
torch~=2.8.0 torch~=2.8.0
PyYAML~=6.0.2 PyYAML~=6.0.2
insightface~=0.7.3

BIN
resources/face/01.png Normal file

Binary file not shown.

After

Width:  |  Height:  |  Size: 2.0 MiB

BIN
resources/face/02.png Normal file

Binary file not shown.

After

Width:  |  Height:  |  Size: 1.6 MiB

BIN
resources/face/16.jpg Normal file

Binary file not shown.

After

Width:  |  Height:  |  Size: 469 KiB