video_detect/service/face_service.py

import cv2
from io import BytesIO
from PIL import Image
from mysql.connector import Error as MySQLError

from ds.db import db

import numpy as np
import threading
from insightface.app import FaceAnalysis

# 全局变量定义
_insightface_app = None
_known_faces_embeddings = {}  # 存储已知人脸特征 {姓名: 特征向量}
_known_faces_names = []  # 存储已知人脸姓名列表


def init_insightface():
    """初始化InsightFace引擎"""
    global _insightface_app
    if _insightface_app is not None:
        print("InsightFace引擎已初始化，无需重复执行")
        return _insightface_app

    try:
        print("正在初始化 InsightFace 引擎（模型：buffalo_l）...")
        # 初始化引擎，指定模型路径和计算 providers
        app = FaceAnalysis(
            name='buffalo_l',
            root='~/.insightface',
            providers=['CPUExecutionProvider']  # 如需GPU可添加'CUDAExecutionProvider'
        )
        app.prepare(ctx_id=0, det_size=(640, 640))  # 调整检测尺寸
        print("InsightFace 引擎初始化完成")

        # 初始化时加载人脸特征库
        init_face_data()

        _insightface_app = app
        return app
    except Exception as e:
        print(f"InsightFace 初始化失败：{str(e)}")
        _insightface_app = None
        return None


def init_face_data():
    """初始化或更新人脸特征库（清空旧数据，避免重复）"""
    global _known_faces_embeddings, _known_faces_names
    # 清空原有数据，防止重复加载
    _known_faces_embeddings.clear()
    _known_faces_names.clear()

    try:
        face_data = get_all_face_name_with_eigenvalue()  # 假设该函数已定义
        print(f"已加载 {len(face_data)} 个人脸数据")
        for person_name, eigenvalue_data in face_data.items():
            # 解析特征值（支持numpy数组或字符串格式）
            if isinstance(eigenvalue_data, np.ndarray):
                eigenvalue = eigenvalue_data.astype(np.float32)
            elif isinstance(eigenvalue_data, str):
                # 增强字符串解析：支持逗号/空格分隔，清理特殊字符
                cleaned = (eigenvalue_data
                           .replace("[", "").replace("]", "")
                           .replace("\n", "").replace(",", " ")
                           .strip())
                values = [v for v in cleaned.split() if v]  # 过滤空字符串
                if not values:
                    print(f"特征值解析失败（空值），跳过 {person_name}")
                    continue
                eigenvalue = np.array(list(map(float, values)), dtype=np.float32)
            else:
                print(f"不支持的特征值类型（{type(eigenvalue_data)}），跳过 {person_name}")
                continue

            # 特征值归一化（确保相似度计算一致性）
            norm = np.linalg.norm(eigenvalue)
            if norm == 0:
                print(f"特征值为零向量，跳过 {person_name}")
                continue
            eigenvalue = eigenvalue / norm

            # 更新全局特征库
            _known_faces_embeddings[person_name] = eigenvalue
            _known_faces_names.append(person_name)

        print(f"成功加载 {len(_known_faces_names)} 个人脸的特征库")
    except Exception as e:
        print(f"加载人脸特征库失败: {e}")


def update_face_data():
    """更新人脸特征库（清空旧数据，加载最新数据）"""
    global _known_faces_embeddings, _known_faces_names

    print("开始更新人脸特征库...")

    # 清空原有数据
    _known_faces_embeddings.clear()
    _known_faces_names.clear()

    try:
        # 获取最新人脸数据
        face_data = get_all_face_name_with_eigenvalue()
        print(f"获取到 {len(face_data)} 条最新人脸数据")

        # 处理并加载新数据（复用原有解析逻辑）
        for person_name, eigenvalue_data in face_data.items():
            # 解析特征值（支持numpy数组或字符串格式）
            if isinstance(eigenvalue_data, np.ndarray):
                eigenvalue = eigenvalue_data.astype(np.float32)
            elif isinstance(eigenvalue_data, str):
                # 增强字符串解析：支持逗号/空格分隔，清理特殊字符
                cleaned = (eigenvalue_data
                           .replace("[", "").replace("]", "")
                           .replace("\n", "").replace(",", " ")
                           .strip())
                values = [v for v in cleaned.split() if v]  # 过滤空字符串
                if not values:
                    print(f"特征值解析失败（空值），跳过 {person_name}")
                    continue
                eigenvalue = np.array(list(map(float, values)), dtype=np.float32)
            else:
                print(f"不支持的特征值类型（{type(eigenvalue_data)}），跳过 {person_name}")
                continue

            # 特征值归一化（确保相似度计算一致性）
            norm = np.linalg.norm(eigenvalue)
            if norm == 0:
                print(f"特征值为零向量，跳过 {person_name}")
                continue
            eigenvalue = eigenvalue / norm

            # 更新全局特征库
            _known_faces_embeddings[person_name] = eigenvalue
            _known_faces_names.append(person_name)

        print(f"人脸特征库更新完成，共加载 {len(_known_faces_names)} 个人脸数据")
        return True  # 更新成功
    except Exception as e:
        print(f"人脸特征库更新失败: {e}")
        return False  # 更新失败


def detect(frame, similarity_threshold=0.4):
    global _insightface_app, _known_faces_embeddings

    # 校验输入有效性
    if frame is None or frame.size == 0:
        return (False, "无效的输入帧数据")

    # 校验引擎和特征库状态
    if not _insightface_app:
        return (False, "人脸引擎未初始化")
    if not _known_faces_embeddings:
        return (False, "人脸特征库为空")

    try:
        # 执行人脸检测与特征提取
        faces = _insightface_app.get(frame)
    except Exception as e:
        return (False, f"检测错误: {str(e)}")

    result_parts = []
    has_matched_known_face = False  # 是否有匹配到已知人脸

    for face in faces:
        # 归一化当前人脸特征
        face_embedding = face.embedding.astype(np.float32)
        norm = np.linalg.norm(face_embedding)
        if norm == 0:
            result_parts.append("检测到人脸但特征值为零向量（忽略）")
            continue
        face_embedding = face_embedding / norm

        # 与已知特征库比对
        max_similarity, best_match_name = -1.0, "Unknown"
        for name, known_emb in _known_faces_embeddings.items():
            similarity = np.dot(face_embedding, known_emb)  # 余弦相似度
            if similarity > max_similarity:
                max_similarity = similarity
                best_match_name = name

        # 判断是否匹配成功
        is_matched = max_similarity >= similarity_threshold

        if is_matched:
            has_matched_known_face = True

        # 记录结果（边界框转为整数列表）
        bbox = face.bbox.astype(int).tolist()
        result_parts.append(
            f"{'匹配' if is_matched else '未匹配'}: {best_match_name} "
            f"(相似度: {max_similarity:.2f}, 边界框: {bbox})"
        )

    # 构建最终结果
    result_str = "未检测到人脸" if not result_parts else "; ".join(result_parts)
    return (has_matched_known_face, result_str)


# 上传图片并提取特征
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
    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:
        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"
            print(error_msg)
            return False, error_msg

    try:
        # 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"
            print(error_msg)
            return False, error_msg

        # 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)}"
            print(error_msg)
            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"
            print(error_msg)
            return False, error_msg

        # 5. 调用InsightFace检测人脸
        print(f"开始检测人脸（图片尺寸：{width}x{height}、格式：BGR）")
        faces = _insightface_app.get(frame)

        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）"
            print(error_msg)
            return False, error_msg

        # 6. 提取特征并保存
        current_feature = faces[0].embedding
        _feature_list.append(current_feature)
        print(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:
        error_msg = f"处理图片时发生异常：{str(e)}"
        print(error_msg)
        return False, error_msg


# 获取数据库最新的人脸及其特征
def get_all_face_name_with_eigenvalue() -> dict:
    conn = None
    cursor = None
    try:
        conn = db.get_connection()
        cursor = conn.cursor(dictionary=True)
        query = "SELECT name, eigenvalue FROM face WHERE name IS NOT NULL"
        cursor.execute(query)
        faces = cursor.fetchall()

        name_to_eigenvalues = {}
        for face in faces:
            name = face["name"]
            eigenvalue = face["eigenvalue"]
            if name in name_to_eigenvalues:
                name_to_eigenvalues[name].append(eigenvalue)
            else:
                name_to_eigenvalues[name] = [eigenvalue]

        face_dict = {}
        for name, eigenvalues in name_to_eigenvalues.items():
            if len(eigenvalues) > 1:
                face_dict[name] = get_average_feature(eigenvalues)
            else:
                face_dict[name] = eigenvalues[0]

        return face_dict

    except MySQLError as e:
        raise Exception(f"获取人脸特征失败: {str(e)}") from e
    finally:
        db.close_connection(conn, cursor)


# 获取平均特征值
def get_average_feature(features=None):
    global _feature_list
    try:
        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")
            return None

        processed_features = []
        for i, embedding in enumerate(features):
            try:
                if isinstance(embedding, str):
                    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)

                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")
                else:
                    print(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：{str(e)}")

        if not processed_features:
            print("没有有效的特征值用于计算平均值")
            return None

        dims = {feat.shape[0] for feat in processed_features}
        if len(dims) > 1:
            print(f"特征值维度不一致：{dims}、无法计算平均值")
            return None

        avg_feature = np.mean(processed_features, axis=0)
        print(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"计算平均特征值出错：{str(e)}")
        return None