目前可以成功动态更换模型运行的
This commit is contained in:
@ -4,6 +4,11 @@ import insightface
|
||||
from insightface.app import FaceAnalysis
|
||||
from io import BytesIO
|
||||
from PIL import Image
|
||||
import logging
|
||||
|
||||
# 配置日志（便于排查）
|
||||
logging.basicConfig(level=logging.INFO, format='%(asctime)s - [FaceUtil] - %(levelname)s - %(message)s')
|
||||
logger = logging.getLogger(__name__)
|
||||
|
||||
# 全局变量存储InsightFace引擎和特征列表
|
||||
_insightface_app = None
|
||||
@ -11,135 +16,141 @@ _feature_list = []
|
||||
|
||||
|
||||
def init_insightface():
|
||||
"""初始化InsightFace引擎"""
|
||||
"""初始化InsightFace引擎（确保成功后再使用）"""
|
||||
global _insightface_app
|
||||
try:
|
||||
print("正在初始化InsightFace引擎...")
|
||||
app = FaceAnalysis(name='buffalo_l', root='~/.insightface')
|
||||
app.prepare(ctx_id=0, det_size=(640, 640))
|
||||
print("InsightFace引擎初始化完成")
|
||||
if _insightface_app is not None:
|
||||
logger.info("InsightFace引擎已初始化，无需重复执行")
|
||||
return _insightface_app
|
||||
|
||||
logger.info("正在初始化InsightFace引擎（模型：buffalo_l）...")
|
||||
# 手动指定模型下载路径（避免权限问题，可选）
|
||||
app = FaceAnalysis(
|
||||
name='buffalo_l',
|
||||
root='~/.insightface', # 模型默认下载路径
|
||||
providers=['CPUExecutionProvider'] # 强制用CPU（若有GPU可加'CUDAExecutionProvider'）
|
||||
)
|
||||
app.prepare(ctx_id=0, det_size=(640, 640)) # det_size越大，小人脸检测越准
|
||||
logger.info("InsightFace引擎初始化完成")
|
||||
_insightface_app = app
|
||||
return app
|
||||
except Exception as e:
|
||||
print(f"InsightFace初始化失败: {e}")
|
||||
logger.error(f"InsightFace初始化失败：{str(e)}", exc_info=True) # 打印详细堆栈
|
||||
_insightface_app = None
|
||||
return None
|
||||
|
||||
|
||||
def add_binary_data(binary_data):
|
||||
"""
|
||||
接收单张图片的二进制数据、提取特征并保存
|
||||
|
||||
参数:
|
||||
binary_data: 图片的二进制数据（bytes类型）
|
||||
|
||||
返回:
|
||||
成功提取特征时返回 (True, 特征值numpy数组)
|
||||
失败时返回 (False, None)
|
||||
返回：(True, 特征值numpy数组) 或 (False, 错误信息字符串)
|
||||
"""
|
||||
global _insightface_app, _feature_list
|
||||
|
||||
# 1. 先检查引擎是否初始化成功
|
||||
if not _insightface_app:
|
||||
print("引擎未初始化、无法处理")
|
||||
return False, None
|
||||
init_result = init_insightface() # 尝试重新初始化
|
||||
if not init_result:
|
||||
error_msg = "InsightFace引擎未初始化，无法检测人脸"
|
||||
logger.error(error_msg)
|
||||
return False, error_msg
|
||||
|
||||
try:
|
||||
# 直接处理二进制数据: 转换为图像格式
|
||||
img = Image.open(BytesIO(binary_data))
|
||||
frame = cv2.cvtColor(np.array(img), cv2.COLOR_RGB2BGR)
|
||||
# 2. 验证二进制数据有效性
|
||||
if len(binary_data) < 1024: # 过滤过小的无效图片（小于1KB）
|
||||
error_msg = f"图片过小（{len(binary_data)}字节），可能不是有效图片"
|
||||
logger.warning(error_msg)
|
||||
return False, error_msg
|
||||
|
||||
# 提取特征
|
||||
# 3. 二进制数据转CV2格式（关键步骤，避免通道错误）
|
||||
try:
|
||||
img = Image.open(BytesIO(binary_data)).convert("RGB") # 强制转RGB
|
||||
frame = cv2.cvtColor(np.array(img), cv2.COLOR_RGB2BGR) # InsightFace需要BGR格式
|
||||
except Exception as e:
|
||||
error_msg = f"图片格式转换失败：{str(e)}"
|
||||
logger.error(error_msg, exc_info=True)
|
||||
return False, error_msg
|
||||
|
||||
# 4. 检查图片尺寸（避免极端尺寸导致检测失败）
|
||||
height, width = frame.shape[:2]
|
||||
if height < 64 or width < 64: # 人脸检测最小建议尺寸
|
||||
error_msg = f"图片尺寸过小（{width}x{height}），需至少64x64像素"
|
||||
logger.warning(error_msg)
|
||||
return False, error_msg
|
||||
|
||||
# 5. 调用InsightFace检测人脸
|
||||
logger.info(f"开始检测人脸（图片尺寸：{width}x{height}，格式：BGR）")
|
||||
faces = _insightface_app.get(frame)
|
||||
if faces:
|
||||
# 获取当前提取的特征值
|
||||
current_feature = faces[0].embedding
|
||||
# 添加到特征列表
|
||||
_feature_list.append(current_feature)
|
||||
print(f"已累计 {len(_feature_list)} 个特征")
|
||||
# 返回成功标志和当前特征值
|
||||
return True, current_feature
|
||||
else:
|
||||
print("二进制数据中未检测到人脸")
|
||||
return False, None
|
||||
|
||||
if not faces:
|
||||
error_msg = "未检测到人脸（请确保图片包含清晰正面人脸，无遮挡、不模糊）"
|
||||
logger.warning(error_msg)
|
||||
return False, error_msg
|
||||
|
||||
# 6. 提取特征并保存
|
||||
current_feature = faces[0].embedding
|
||||
_feature_list.append(current_feature)
|
||||
logger.info(f"人脸检测成功，提取特征值（维度：{current_feature.shape[0]}），累计特征数：{len(_feature_list)}")
|
||||
return True, current_feature
|
||||
|
||||
except Exception as e:
|
||||
print(f"处理二进制数据出错: {e}")
|
||||
return False, None
|
||||
error_msg = f"处理图片时发生异常：{str(e)}"
|
||||
logger.error(error_msg, exc_info=True)
|
||||
return False, error_msg
|
||||
|
||||
|
||||
# 以下函数保持不变（get_average_feature/clear_features/get_feature_list）
|
||||
def get_average_feature(features=None):
|
||||
"""
|
||||
计算多个特征向量的平均值
|
||||
|
||||
参数:
|
||||
features: 可选、特征值列表。如果未提供、则使用全局存储的_feature_list
|
||||
每个元素可以是字符串格式或numpy数组
|
||||
|
||||
返回:
|
||||
单一平均特征向量的numpy数组、若无可计算数据则返回None
|
||||
"""
|
||||
global _feature_list
|
||||
|
||||
# 如果未提供features参数、则使用全局特征列表
|
||||
if features is None:
|
||||
features = _feature_list
|
||||
|
||||
try:
|
||||
# 验证输入是否为列表且不为空
|
||||
if features is None:
|
||||
features = _feature_list
|
||||
if not isinstance(features, list) or len(features) == 0:
|
||||
print("输入必须是包含至少一个特征值的列表")
|
||||
logger.warning("输入必须是包含至少一个特征值的列表")
|
||||
return None
|
||||
|
||||
# 处理每个特征值
|
||||
processed_features = []
|
||||
for i, embedding in enumerate(features):
|
||||
try:
|
||||
if isinstance(embedding, str):
|
||||
# 处理包含括号和逗号的字符串格式
|
||||
embedding_str = embedding.replace('[', '').replace(']', '').replace(',', ' ').strip()
|
||||
embedding_list = [float(num) for num in embedding_str.split() if num.strip()]
|
||||
embedding_np = np.array(embedding_list, dtype=np.float32)
|
||||
else:
|
||||
embedding_np = np.array(embedding, dtype=np.float32)
|
||||
|
||||
# 验证特征值格式
|
||||
if len(embedding_np.shape) == 1:
|
||||
processed_features.append(embedding_np)
|
||||
print(f"已添加第 {i + 1} 个特征值用于计算平均值")
|
||||
logger.info(f"已添加第 {i + 1} 个特征值用于计算平均值")
|
||||
else:
|
||||
print(f"跳过第 {i + 1} 个特征值、不是一维数组")
|
||||
|
||||
logger.warning(f"跳过第 {i + 1} 个特征值：不是一维数组")
|
||||
except Exception as e:
|
||||
print(f"处理第 {i + 1} 个特征值时出错: {e}")
|
||||
logger.error(f"处理第 {i + 1} 个特征值时出错：{str(e)}")
|
||||
|
||||
# 确保有有效的特征值
|
||||
if not processed_features:
|
||||
print("没有有效的特征值用于计算平均值")
|
||||
logger.warning("没有有效的特征值用于计算平均值")
|
||||
return None
|
||||
|
||||
# 检查所有特征向量维度是否相同
|
||||
dims = {feat.shape[0] for feat in processed_features}
|
||||
if len(dims) > 1:
|
||||
print(f"特征值维度不一致、无法计算平均值。检测到的维度: {dims}")
|
||||
logger.error(f"特征值维度不一致：{dims}，无法计算平均值")
|
||||
return None
|
||||
|
||||
# 计算平均值
|
||||
avg_feature = np.mean(processed_features, axis=0)
|
||||
print(f"成功计算 {len(processed_features)} 个特征值的平均特征向量、维度: {avg_feature.shape[0]}")
|
||||
|
||||
logger.info(f"计算成功：{len(processed_features)} 个特征值的平均向量（维度：{avg_feature.shape[0]}）")
|
||||
return avg_feature
|
||||
|
||||
except Exception as e:
|
||||
print(f"计算平均特征值时出错: {e}")
|
||||
logger.error(f"计算平均特征值出错：{str(e)}", exc_info=True)
|
||||
return None
|
||||
|
||||
|
||||
def clear_features():
|
||||
"""清空已存储的特征数据"""
|
||||
global _feature_list
|
||||
_feature_list = []
|
||||
print("已清空所有特征数据")
|
||||
logger.info("已清空所有特征数据")
|
||||
|
||||
|
||||
def get_feature_list():
|
||||
"""获取当前存储的特征列表"""
|
||||
global _feature_list
|
||||
return _feature_list.copy() # 返回副本防止外部直接修改
|
||||
logger.info(f"当前特征列表长度：{len(_feature_list)}")
|
||||
return _feature_list.copy()
|
||||
83
util/file_util.py
Normal file
83
util/file_util.py
Normal file
@ -0,0 +1,83 @@
|
||||
import os
|
||||
import datetime
|
||||
from pathlib import Path
|
||||
from typing import Dict
|
||||
|
||||
|
||||
def save_face_to_up_images(
|
||||
client_ip: str,
|
||||
face_name: str,
|
||||
image_bytes: bytes,
|
||||
image_format: str = "jpg"
|
||||
) -> Dict[str, str]:
|
||||
"""
|
||||
保存人脸图片到 `/up_images/用户IP/人脸名字/` 路径
|
||||
修复路径计算错误，确保所有路径在up_images根目录下
|
||||
|
||||
参数:
|
||||
client_ip: 客户端IP（原始格式，如192.168.1.101）
|
||||
face_name: 人脸名称（用户输入，可为空）
|
||||
image_bytes: 人脸图片二进制数据
|
||||
image_format: 图片格式（默认jpg）
|
||||
|
||||
返回:
|
||||
字典：success（是否成功）、db_path（存数据库的相对路径）、local_abs_path（本地绝对路径）、msg（提示）
|
||||
"""
|
||||
try:
|
||||
# 1. 基础参数校验
|
||||
if not client_ip.strip():
|
||||
return {"success": False, "db_path": "", "local_abs_path": "", "msg": "客户端IP不能为空"}
|
||||
if not image_bytes:
|
||||
return {"success": False, "db_path": "", "local_abs_path": "", "msg": "图片二进制数据为空"}
|
||||
if image_format.lower() not in ["jpg", "jpeg", "png"]:
|
||||
return {"success": False, "db_path": "", "local_abs_path": "", "msg": "仅支持jpg/jpeg/png格式"}
|
||||
|
||||
# 2. 处理特殊字符（避免路径错误）
|
||||
safe_ip = client_ip.strip().replace(".", "_") # IP中的.替换为_
|
||||
safe_face_name = face_name.strip() if (face_name and face_name.strip()) else "未命名"
|
||||
safe_face_name = "".join([c for c in safe_face_name if c not in r'\/:*?"<>|']) # 过滤非法字符
|
||||
|
||||
# 3. 构建根目录（强制转为绝对路径，避免相对路径混淆）
|
||||
root_dir = Path("up_images").resolve() # 转为绝对路径（关键修复！）
|
||||
if not root_dir.exists():
|
||||
root_dir.mkdir(parents=True, exist_ok=True)
|
||||
print(f"[FileUtil] 已创建up_images根目录：{root_dir}")
|
||||
|
||||
# 4. 构建文件层级路径（确保在root_dir子目录下）
|
||||
ip_dir = root_dir / safe_ip
|
||||
face_name_dir = ip_dir / safe_face_name
|
||||
face_name_dir.mkdir(parents=True, exist_ok=True) # 自动创建目录
|
||||
print(f"[FileUtil] 图片存储目录：{face_name_dir}")
|
||||
|
||||
# 5. 生成唯一文件名（毫秒级时间戳）
|
||||
timestamp = datetime.datetime.now().strftime("%Y%m%d%H%M%S%f")[:-3]
|
||||
image_filename = f"face_{safe_ip}_{timestamp}.{image_format.lower()}"
|
||||
|
||||
# 6. 计算路径（关键修复：确保所有路径都是绝对路径且在root_dir下）
|
||||
local_abs_path = face_name_dir / image_filename # 绝对路径
|
||||
|
||||
# 验证路径是否在root_dir下（防止路径穿越攻击）
|
||||
if not local_abs_path.resolve().is_relative_to(root_dir.resolve()):
|
||||
raise Exception(f"图片路径不在up_images根目录下（安全校验失败）：{local_abs_path}")
|
||||
|
||||
# 数据库存储路径：从root_dir开始的相对路径（如 up_images/192_168_110_31/小王/xxx.jpg）
|
||||
db_path = str(root_dir.name / local_abs_path.relative_to(root_dir))
|
||||
|
||||
# 7. 写入图片文件
|
||||
with open(local_abs_path, "wb") as f:
|
||||
f.write(image_bytes)
|
||||
print(f"[FileUtil] 图片保存成功：")
|
||||
print(f" 数据库路径：{db_path}")
|
||||
print(f" 本地绝对路径：{local_abs_path}")
|
||||
|
||||
return {
|
||||
"success": True,
|
||||
"db_path": db_path, # 存数据库的相对路径（up_images开头）
|
||||
"local_abs_path": str(local_abs_path), # 本地绝对路径
|
||||
"msg": "图片保存成功"
|
||||
}
|
||||
|
||||
except Exception as e:
|
||||
error_msg = f"图片保存失败：{str(e)}"
|
||||
print(f"[FileUtil] 错误：{error_msg}")
|
||||
return {"success": False, "db_path": "", "local_abs_path": "", "msg": error_msg}
|
||||
61
util/model_util.py
Normal file
61
util/model_util.py
Normal file
@ -0,0 +1,61 @@
|
||||
import os
|
||||
import numpy as np
|
||||
import traceback
|
||||
from ultralytics import YOLO
|
||||
from typing import Optional
|
||||
|
||||
|
||||
def load_yolo_model(model_path: str) -> Optional[YOLO]:
|
||||
"""
|
||||
加载YOLO模型（支持v5/v8），并校验模型有效性
|
||||
:param model_path: 模型文件的绝对路径
|
||||
:return: 加载成功返回YOLO模型实例，失败返回None
|
||||
"""
|
||||
try:
|
||||
# 加载前的基础信息检查
|
||||
print(f"\n[模型工具] 开始加载模型：{model_path}")
|
||||
print(f"[模型工具] 文件是否存在：{os.path.exists(model_path)}")
|
||||
if os.path.exists(model_path):
|
||||
print(f"[模型工具] 文件大小：{os.path.getsize(model_path) / 1024 / 1024:.2f} MB")
|
||||
|
||||
# 强制重新加载模型，避免缓存问题
|
||||
model = YOLO(model_path)
|
||||
|
||||
# 兼容性校验：使用numpy空数组测试模型
|
||||
dummy_image = np.zeros((640, 640, 3), dtype=np.uint8)
|
||||
|
||||
try:
|
||||
# 优先使用新版本参数
|
||||
model.predict(
|
||||
source=dummy_image,
|
||||
imgsz=640,
|
||||
conf=0.25,
|
||||
verbose=False,
|
||||
stream=False
|
||||
)
|
||||
except Exception as pred_e:
|
||||
print(f"[模型工具] 预测校验兼容处理：{str(pred_e)}")
|
||||
# 兼容旧版本YOLO参数
|
||||
model.predict(
|
||||
img=dummy_image,
|
||||
imgsz=640,
|
||||
conf=0.25,
|
||||
verbose=False
|
||||
)
|
||||
|
||||
# 验证模型基本属性
|
||||
if not hasattr(model, 'names'):
|
||||
print("[模型工具] 警告：模型缺少类别名称属性")
|
||||
else:
|
||||
print(f"[模型工具] 模型包含类别：{list(model.names.values())[:5]}...") # 显示前5个类别
|
||||
|
||||
print(f"[模型工具] 模型加载成功！")
|
||||
return model
|
||||
|
||||
except Exception as e:
|
||||
# 详细错误信息输出
|
||||
print(f"\n[模型工具] 加载模型失败！路径：{model_path}")
|
||||
print(f"[模型工具] 异常类型：{type(e).__name__}")
|
||||
print(f"[模型工具] 异常详情：{str(e)}")
|
||||
print(f"[模型工具] 堆栈跟踪：\n{traceback.format_exc()}")
|
||||
return None
|
||||
Reference in New Issue
Block a user