Yolov/detectionThread.py

import datetime
import gc
import json
import os
import queue
import threading
import time
import traceback

import cv2
import paho.mqtt.client as mqtt
import requests
import torch
from ultralytics import YOLO

from _minio import MinioUploader
from log import logger
from global_data import gd
from detection_render import OptimizedDetectionRenderer


class ModelManager:
    """模型管理器，支持多模型和加密模型"""

    def __init__(self, config):
        self.config = config
        self.models_dir = "models"
        os.makedirs(self.models_dir, exist_ok=True)

    def load_model(self, model_config, require_verification=False):
        """加载单个模型（支持加密），可选密钥验证"""
        try:
            model_path = model_config['path']
            encrypted = model_config.get('encrypted', False)
            encryption_key = model_config.get('encryption_key')

            # 构建本地路径
            local_path = os.path.join(self.models_dir, os.path.basename(model_path))

            # 下载模型（如果不存在且提供下载地址）
            if not os.path.exists(local_path):
                download_url = model_config.get('download_url')
                if download_url:
                    if not self.download_model(model_config, local_path):
                        logger.warning(f"模型文件不存在且下载失败: {local_path}")
                        return None, None  # 返回模型和验证结果
                else:
                    logger.warning(f"模型文件不存在: {local_path}")
                    return None, None

            # 验证加密模型密钥（如果需要）
            verification_result = None
            if encrypted and encryption_key:
                if require_verification:
                    verification_result = self.verify_model_key(local_path, encryption_key)
                    if not verification_result.get('success', False):
                        logger.error(f"加密模型密钥验证失败: {model_path}")
                        return None, verification_result
                    logger.info(f"加密模型密钥验证成功: {model_path}")

                # 解密并加载模型
                try:
                    from mandatory_model_crypto import MandatoryModelEncryptor
                    encryptor = MandatoryModelEncryptor()

                    # 解密模型
                    decrypt_result = encryptor.decrypt_model(local_path, encryption_key,
                                                             verify_key=require_verification)

                    if not decrypt_result.get('success', False):
                        logger.error(f"解密模型失败: {model_path}")
                        return None, decrypt_result

                    decrypted_path = decrypt_result.get('temp_path')
                    if not decrypted_path:
                        logger.error(f"未获取到解密后的模型路径: {model_path}")
                        return None, decrypt_result

                    logger.info(f"解密模型成功: {model_path}")
                    model = YOLO(decrypted_path)

                    # 清理临时文件
                    try:
                        os.unlink(decrypted_path)
                    except Exception as e:
                        logger.warning(f"清理临时文件失败: {str(e)}")

                except ImportError:
                    logger.error("mandatory_model_crypto模块未找到，无法处理加密模型")
                    return None, {'success': False, 'error': '加密模块未找到'}
                except Exception as e:
                    logger.error(f"加密模型处理失败: {str(e)}")
                    return None, {'success': False, 'error': str(e)}
            else:
                # 普通模型加载
                if encrypted and not encryption_key:
                    logger.warning(f"模型标记为加密但未提供密钥: {model_path}")
                model = YOLO(local_path)
                verification_result = {'success': True} if not encrypted else {
                    'success': False,
                    'error': '模型标记为加密但未提供密钥'
                }

            # 应用设备配置
            device = model_config.get('device', 'cuda:0' if torch.cuda.is_available() else 'cpu')
            model = model.to(device)

            # 应用半精度配置
            if model_config.get('half', False) and 'cuda' in device:
                model = model.half()
                logger.info(f"启用半精度推理: {model_path}")

            logger.info(f"模型加载成功: {model_path} -> {device}")
            return model, verification_result

        except Exception as e:
            logger.error(f"加载模型失败 {model_config.get('path')}: {str(e)}")
            logger.error(traceback.format_exc())
            return None, {'success': False, 'error': str(e)}

    def download_model(self, model_config, save_path):
        """下载模型文件"""
        try:
            download_url = model_config.get('download_url')

            if not download_url:
                logger.error(f"模型无下载地址: {model_config['path']}")
                return False

            logger.info(f"下载模型: {download_url} -> {save_path}")

            response = requests.get(download_url, stream=True, timeout=30)
            response.raise_for_status()

            total_size = int(response.headers.get('content-length', 0))
            downloaded = 0

            with open(save_path, 'wb') as f:
                for chunk in response.iter_content(chunk_size=8192):
                    if chunk:
                        downloaded += len(chunk)
                        f.write(chunk)

                        if total_size > 0:
                            progress = (downloaded * 100) // total_size
                            if progress % 25 == 0:  # 每25%输出一次
                                logger.info(f"下载进度: {progress}%")

            logger.info(f"模型下载完成: {save_path} ({downloaded} 字节)")
            return True

        except Exception as e:
            logger.error(f"下载模型失败: {str(e)}")
            return False

    def verify_model_key(self, path, encryption_key):
        """验证模型密钥"""
        try:
            from mandatory_model_crypto import MandatoryModelEncryptor
            encryptor = MandatoryModelEncryptor()
            return encryptor.decrypt_model(path, encryption_key, verify_key=True)
        except ImportError:
            return {'success': False, 'error': '加密模块未找到'}
        except Exception as e:
            return {'success': False, 'error': str(e)}


class DetectionThread(threading.Thread):
    """多模型检测线程 - 优化版本"""

    def __init__(self, config):
        super().__init__()
        self.config = config
        self.task_id = None
        self.initialized = False
        self.running = True
        self._cleaning_up = False
        self._force_stop = False
        self._should_stop = threading.Event()  # 添加停止事件

        # 多模型支持
        self.models = []  # 存储多个模型及相关配置
        self.model_manager = ModelManager(config)
        self.key_verification_results = {}  # 密钥验证结果

        # RTMP配置
        self.cap = None
        self.rtmp_url = config['rtmp']['url']
        self.max_reconnect_attempts = config['rtmp']['max_reconnect_attempts']
        self.reconnect_delay = config['rtmp']['reconnect_delay']
        self.buffer_size = config['rtmp']['buffer_size']
        self.timeout_ms = config['rtmp']['timeout_ms']

        # 流稳定性监控
        self.stream_stable = True
        self.consecutive_read_failures = 0
        self.max_consecutive_failures = 20  # 连续读取失败最大次数
        self.retry_delay = 0.5  # 读取失败后的重试延迟（秒）
        self.stream_check_interval = 10  # 流稳定性检查间隔（秒）
        self.last_stream_check = time.time()
        self.total_read_failures = 0
        self.stream_recovery_attempts = 0

        # 任务信息
        self.taskname = config['task']['taskname']
        self.taskid = config['task']['taskid']
        self.aiid = config['task']['aiid']

        # 推流管理
        self.enable_push = config['push']['enable_push']
        self.push_config = config['push']
        self.task_streamer = None  # 任务独立的推流器
        self.streamer_initialized = False
        self.last_push_time = 0
        self.push_error_count = 0
        self.max_push_errors = 5

        # 性能监控
        self.frame_count = 0
        self.reconnect_attempts = 0
        self.last_frame_time = time.time()
        self.fps = 0
        self.detections_count = 0
        self.stop_event = threading.Event()
        self.last_log_time = time.time()
        self.daemon = True
        self.target_latency = 0.05
        self.processing_times = []
        self.avg_process_time = 0.033
        self.last_status_update = 0
        self.last_fps = 0
        self.original_width = 0
        self.original_height = 0

        # 上传配置
        self.minio_uploader = MinioUploader(config['minio'])
        self.upload_queue = queue.Queue(maxsize=50)
        self.upload_thread = None
        self.upload_active = False
        self.upload_interval = 2
        self.last_upload_time = 0
        self.res_api = config['task']['res_api']

        # MQTT配置
        self.mqtt_config = config.get('mqtt', {})
        self.mqtt_enabled = self.mqtt_config.get('enable', False)
        self.mqtt_topic = self.mqtt_config.get('topic', 'drone/data')
        self.mqtt_client = None
        self.mqtt_connected = False
        self.latest_drone_data = None
        self.mqtt_data_lock = threading.Lock()

        # 颜色配置
        self.color_palette = [
            (0, 255, 0),  # 绿色
            (255, 0, 0),  # 蓝色
            (0, 0, 255),  # 红色
            (255, 255, 0),  # 青色
            (255, 0, 255),  # 紫色
            (0, 255, 255),  # 黄色
            (128, 0, 128),  # 深紫色
            (0, 128, 128),  # 深青色
            (128, 128, 0),  # 橄榄色
        ]
        # Windows推流管理
        if self._is_windows():
            from windows_utils import detect_and_configure_windows
            from task_stream_manager_windows import windows_task_stream_manager

            # 检测和配置Windows
            self.windows_config = detect_and_configure_windows()
            self.stream_manager = windows_task_stream_manager
            logger.info(f"Windows系统检测完成: {self.windows_config.get('status', 'unknown')}")
        else:
            # Linux/Mac使用原管理器
            from task_stream_manager import task_stream_manager
            self.stream_manager = task_stream_manager
            # 推流状态跟踪
        self.stream_stats = {
            'total_frames_pushed': 0,
            'failed_pushes': 0,
            'last_push_time': 0,
            'push_success_rate': 1.0,
            'ffmpeg_restarts': 0
        }
        # 密钥验证记录
        self.key_verification_results = {}

        # 绘制结果
        self.renderer = OptimizedDetectionRenderer()
        logger.info(f"检测线程初始化完成: {self.taskname}")

    def _is_windows(self):
        """检查是否是Windows系统"""
        import os
        return os.name == 'nt' or os.name == 'win32'

    def load_models(self):
        """加载多个模型 - 简化版本"""
        try:
            models_config = self.config.get('models', [])
            if not models_config or not isinstance(models_config, list):
                logger.error("未找到有效的models配置列表")
                return False

            logger.info(f"开始加载 {len(models_config)} 个模型")

            loaded_models = []
            key_verification_results = {}

            for i, model_config in enumerate(models_config):
                # 检查模型是否启用
                if not model_config.get('enabled', True):
                    logger.info(f"跳过未启用的模型 {i}")
                    continue

                model_path = model_config.get('path', 'unknown')
                model_name = os.path.basename(model_path).split('.')[0]

                # 强制验证加密模型密钥
                require_verification = model_config.get('encrypted', False)

                # 加载模型（包含密钥验证）
                logger.info(f"加载模型 {i}: {model_name}")
                model, verification_result = self.model_manager.load_model(
                    model_config,
                    require_verification=require_verification
                )

                # 记录验证结果
                key_verification_results[i] = verification_result or {'success': False, 'error': '未知错误'}

                if model is None:
                    logger.error(f"加载模型 {i} 失败: {model_name}")
                    continue

                # 为模型分配颜色
                model_color = model_config.get('color')
                if model_color is None:
                    color_idx = i % len(self.color_palette)
                    model_color = self.color_palette[color_idx]

                # 准备标签
                tags = model_config.get('tags', {})

                # 存储模型信息
                model_info = {
                    'model': model,
                    'config': model_config,
                    'tags': tags,
                    'color': model_color,
                    'name': model_name,
                    'id': i,
                    'device': model_config.get('device', 'cpu'),
                    'imgsz': model_config.get('imgsz', 640),
                    'conf_thres': model_config.get('conf_thres', 0.25),
                    'iou_thres': model_config.get('iou_thres', 0.45),
                    'half': model_config.get('half', False),
                    'key_valid': verification_result.get('success', False) if verification_result else False,
                    'model_hash': verification_result.get('model_hash', 'unknown') if verification_result else 'unknown'
                }

                loaded_models.append(model_info)
                logger.info(
                    f"模型加载成功: {model_name}, 颜色: {model_color}, 设备: {model_info['device']}, 密钥验证: {model_info['key_valid']}")

            # 检查加载结果
            if len(loaded_models) == 0:
                logger.error("所有模型加载失败")

                # 生成详细的错误报告
                error_report = []
                for idx, result in key_verification_results.items():
                    if not result.get('success', False):
                        error_report.append(f"模型 {idx}: {result.get('error', '未知错误')}")

                if error_report:
                    logger.error("模型加载失败详情:")
                    for error in error_report:
                        logger.error(f"  {error}")

                return False

            self.models = loaded_models
            self.key_verification_results = key_verification_results

            logger.info(f"成功加载 {len(self.models)}/{len(models_config)} 个模型")

            # 输出验证统计
            valid_keys = sum(1 for result in key_verification_results.values()
                             if result.get('success', False))
            logger.info(f"密钥验证统计: {valid_keys}个有效, {len(key_verification_results) - valid_keys}个无效")

            return True

        except Exception as e:
            logger.error(f"加载模型异常: {str(e)}")
            logger.error(traceback.format_exc())
            return False

    def get_key_verification_summary(self):
        """获取密钥验证摘要"""
        summary = {
            'total_models': len(self.key_verification_results),
            'valid_keys': 0,
            'invalid_keys': 0,
            'loaded_models': len(self.models),
            'details': {}
        }

        for idx, result in self.key_verification_results.items():
            if result.get('success', False):
                summary['valid_keys'] += 1
            else:
                summary['invalid_keys'] += 1

            summary['details'][f'model_{idx}'] = {
                'success': result.get('success', False),
                'error': result.get('error', ''),
                'model_hash': result.get('model_hash', '')
            }

        return summary

    def initialize_rtmp(self):
        """初始化RTMP连接"""
        try:
            logger.info(f"连接RTMP: {self.rtmp_url}")
            self.cap = cv2.VideoCapture()

            # 设置优化参数
            self.cap.set(cv2.CAP_PROP_BUFFERSIZE, self.buffer_size)
            self.cap.set(cv2.CAP_PROP_FOURCC, cv2.VideoWriter_fourcc(*'H264'))
            self.cap.set(cv2.CAP_PROP_FPS, 30)

            # 启用硬件加速解码
            if self.config['rtmp'].get('gpu_decode', False):
                try:
                    self.cap.set(cv2.CAP_PROP_HW_ACCELERATION, cv2.VIDEO_ACCELERATION_ANY)
                    logger.info("启用硬件加速解码")
                except:
                    logger.warning("硬件解码不可用，使用软件解码")

            # 尝试连接
            if not self.cap.open(self.rtmp_url, cv2.CAP_FFMPEG):
                logger.error(f"连接RTMP失败: {self.rtmp_url}")
                raise IOError(f"无法连接RTMP流 ({self.rtmp_url})")

            # 获取视频属性
            self.original_width = int(self.cap.get(cv2.CAP_PROP_FRAME_WIDTH))
            self.original_height = int(self.cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
            fps = self.cap.get(cv2.CAP_PROP_FPS) or 30
            self.fps = fps

            # 重置流稳定性计数器
            self.consecutive_read_failures = 0
            self.total_read_failures = 0
            self.stream_stable = True

            logger.info(f"视频属性: {self.original_width}x{self.original_height} @ {fps}fps")
            return True

        except Exception as e:
            logger.error(f"初始化RTMP失败: {str(e)}")
            raise

    def initialize_task_streamer(self):
        """初始化任务推流器（Windows优化版）"""
        if not self.enable_push:
            logger.info(f"任务 {self.task_id} 推流功能未启用")
            return True

        try:
            logger.info(f"初始化任务推流器: {self.task_id}")

            # Windows特定检查和配置
            if self._is_windows():
                # 检查RTMP服务器可达性
                from windows_utils import WindowsSystemUtils
                push_url = self.push_config.get('url', '')
                if push_url:
                    accessibility = WindowsSystemUtils.check_rtmp_server_accessibility(push_url)
                    if not accessibility.get('accessible', False):
                        logger.error(f"RTMP服务器不可达: {accessibility.get('error', 'Unknown error')}")
                        logger.warning("推流可能失败，请检查网络和服务器状态")

            # 使用推流管理器创建推流器
            streamer = self.stream_manager.create_streamer_for_task(
                self.task_id,
                self.config,  # 传递完整配置
                self.fps,
                self.original_width,
                self.original_height
            )

            if streamer:
                logger.info(f"任务 {self.task_id} 推流器初始化成功")
                self.streamer_initialized = True
                self.task_streamer = streamer

                # 记录初始化时间
                self.stream_stats['streamer_start_time'] = time.time()

                return True
            else:
                logger.error(f"任务 {self.task_id} 推流器初始化失败")

                # Windows上尝试备用配置
                if self._is_windows():
                    logger.info("尝试Windows备用推流配置...")
                    return self._initialize_windows_fallback_streamer()

                return False

        except Exception as e:
            logger.error(f"初始化任务推流器异常 {self.task_id}: {str(e)}", exc_info=True)

            # Windows上的额外错误处理
            if self._is_windows():
                return self._initialize_windows_fallback_streamer()

            return False

    def _initialize_windows_fallback_streamer(self):
        """Windows备用推流器初始化"""
        try:
            logger.info(f"使用Windows备用推流配置: {self.task_id}")

            # 简化配置，避免硬件加速
            fallback_config = self.config.copy()
            fallback_config['push'] = {
                'enable_push': True,
                'url': self.push_config['url'],
                'video_codec': 'libx264',
                'preset': 'ultrafast',
                'tune': 'zerolatency',
                'format': 'flv',
                'pixel_format': 'bgr24',
                'gpu_acceleration': False,
                'bitrate': '1000k',
                'bufsize': '2000k',
                'framerate': self.fps,
                'extra_args': [
                    '-max_delay', '0',
                    '-flags', '+global_header',
                    '-rtbufsize', '50M'
                ]
            }

            # 使用简化配置重新初始化
            streamer = self.stream_manager.create_streamer_for_task(
                self.task_id,
                fallback_config,
                self.fps,
                self.original_width,
                self.original_height
            )

            if streamer:
                logger.info(f"Windows备用推流器初始化成功: {self.task_id}")
                self.streamer_initialized = True
                self.task_streamer = streamer
                return True
            else:
                logger.error(f"Windows备用推流器也失败: {self.task_id}")
                return False

        except Exception as e:
            logger.error(f"Windows备用推流器初始化异常: {str(e)}")
            return False

    def push_frame_to_task_streamer(self, frame):
        """推送帧到任务推流器（Windows优化版）"""
        if not self.enable_push or not self.streamer_initialized:
            return False

        try:
            # 使用推流管理器推送帧
            success = self.stream_manager.push_frame(self.task_id, frame)

            # 更新统计信息
            self.stream_stats['total_frames_pushed'] += 1

            if success:
                self.last_push_time = time.time()
                self.stream_stats['last_push_time'] = time.time()
                self.push_error_count = 0

                # 更新成功率
                total = self.stream_stats['total_frames_pushed']
                failed = self.stream_stats['failed_pushes']
                self.stream_stats['push_success_rate'] = (total - failed) / max(total, 1)

                # 每100帧记录一次成功
                if total % 100 == 0:
                    logger.info(
                        f"任务 {self.task_id} 已成功推流 {total} 帧，成功率: {self.stream_stats['push_success_rate']:.2%}")
            else:
                self.push_error_count += 1
                self.stream_stats['failed_pushes'] += 1

                logger.warning(f"任务 {self.task_id} 推流失败 ({self.push_error_count}/{self.max_push_errors})")

                # Windows上的额外诊断
                if self._is_windows() and self.push_error_count >= 3:
                    self._diagnose_windows_streaming_issue()

                # 连续失败处理
                if self.push_error_count >= self.max_push_errors:
                    logger.error(f"任务 {self.task_id} 推流连续失败，尝试恢复")
                    self.recover_task_streamer()

            return success

        except Exception as e:
            logger.error(f"推流异常 {self.task_id}: {str(e)}")
            self.push_error_count += 1
            self.stream_stats['failed_pushes'] += 1
            return False

    def _diagnose_windows_streaming_issue(self):
        """诊断Windows推流问题"""
        try:
            logger.info("诊断Windows推流问题...")

            # 获取推流信息
            stream_info = self.stream_manager.get_task_stream_info(self.task_id)
            if stream_info:
                logger.info(f"推流状态: {stream_info.get('status', 'unknown')}")
                logger.info(f"FFmpeg进程: {'运行中' if stream_info.get('process_alive') else '已停止'}")
                logger.info(f"最近输出: {stream_info.get('last_ffmpeg_output', '无')}")

                # 显示最近错误
                output_lines = stream_info.get('output_lines', [])
                if output_lines:
                    logger.info("最近FFmpeg输出:")
                    for line in output_lines[-5:]:
                        logger.info(f"  {line}")

            # 检查系统资源
            from windows_utils import WindowsSystemUtils
            resources = WindowsSystemUtils.get_system_resources()

            if resources['cpu_percent'] > 90:
                logger.warning(f"CPU使用率过高: {resources['cpu_percent']}%")
            if resources['memory_percent'] > 90:
                logger.warning(f"内存使用率过高: {resources['memory_percent']}%")

        except Exception as e:
            logger.error(f"诊断推流问题异常: {str(e)}")

    def recover_task_streamer(self):
        """恢复任务推流器（Windows优化版）"""
        try:
            logger.info(f"任务 {self.task_id} 尝试恢复推流器")

            # Windows上增加重启计数
            self.stream_stats['ffmpeg_restarts'] += 1

            # 清理并重新初始化
            self.cleanup_task_streamer()
            time.sleep(2)  # Windows上需要更长的等待时间

            success = self.initialize_task_streamer()
            if success:
                self.push_error_count = 0
                logger.info(f"任务 {self.task_id} 推流器恢复成功 (第{self.stream_stats['ffmpeg_restarts']}次重启)")
            else:
                logger.error(f"任务 {self.task_id} 推流器恢复失败")

            return success

        except Exception as e:
            logger.error(f"恢复推流器异常 {self.task_id}: {str(e)}")
            return False

    def cleanup_task_streamer(self):
        """清理任务推流器"""
        if self.task_streamer:
            try:
                self.task_streamer.stop()
                self.stream_manager.stop_task_streamer(self.task_id)
                self.task_streamer = None
                logger.info(f"任务 {self.task_id} 推流器已清理")
            except Exception as e:
                logger.error(f"停止推流器失败: {str(e)}")
        self.streamer_initialized = False

    def warmup_models(self):
        """预热所有模型"""
        logger.info("预热所有模型...")
        for model_info in self.models:
            model = model_info['model']
            model_config = model_info['config']

            try:
                # 准备预热输入
                imgsz = model_config.get('imgsz', 640)
                dummy_input = torch.zeros(1, 3, imgsz, imgsz)
                device = model_config.get('device', 'cpu')
                dummy_input = dummy_input.to(device)

                if model_config.get('half', False) and 'cuda' in device:
                    dummy_input = dummy_input.half()

                # 预热推理
                with torch.no_grad():
                    for _ in range(2):
                        model.predict(dummy_input)

                logger.info(f"模型 {model_info['name']} 预热完成")
            except Exception as e:
                logger.warning(f"模型 {model_info['name']} 预热失败: {str(e)}")

        logger.info("所有模型预热完成")

    def multi_model_inference(self, frame):
        """多模型推理（每个模型独立标签和置信度）"""
        all_detections = {}
        annotated_frame = frame.copy()
        model_config_list = {}
        model_detections_list = []
        for model_info in self.models:
            model = model_info['model']
            model_config = model_info['config']
            model_tags = model_info['tags']  # 模型独立标签
            model_color = model_info['color']
            model_name = model_info['name']
            model_id = model_info['id']
            model_config_list[model_id] = model_config
            try:
                # 使用模型特定的参数
                conf_thres = model_config.get('conf_thres', 0.25)
                iou_thres = model_config.get('iou_thres', 0.45)
                imgsz = model_config.get('imgsz', 640)
                device = model_config.get('device', 'cpu')
                half = model_config.get('half', False) and 'cuda' in device

                with torch.no_grad():
                    results = model.predict(
                        frame,
                        stream=False,
                        verbose=False,
                        conf=conf_thres,
                        iou=iou_thres,
                        imgsz=imgsz,
                        device=device,
                        half=half
                    )

                # 处理结果
                result = results[0]
                detections = []

                if result.boxes is not None:
                    for box in result.boxes:
                        class_id = int(box.cls)
                        confidence = float(box.conf)

                        # 从模型特定标签中获取类别信息
                        class_key = str(class_id)
                        class_info = model_tags.get(class_key, None)

                        if class_info is None:
                            continue

                        # 使用标签中的reliability作为置信度阈值
                        tag_reliability = class_info.get('reliability', conf_thres)
                        if confidence < tag_reliability:
                            continue

                        detection_info = {
                            'model_id': model_id,
                            'model_name': model_name,
                            'class_id': class_id,
                            'class_name': class_info.get('name', f'class_{class_id}'),
                            'confidence': confidence,
                            'box': box.xyxy[0].tolist(),
                            'reliability': tag_reliability,
                            'color': model_color
                        }

                        detections.append(detection_info)
                        model_detections_list.append(detection_info)

                all_detections[model_name] = detections

            except Exception as e:
                logger.error(f"模型 {model_name} 推理失败: {str(e)}")
                all_detections[model_name] = []
        self.renderer.draw_all_detections(
            frame,
            model_detections_list,
            model_config_list,
            enable_nms=True,  # 启用NMS去重
            show_model_name=True
        )
        return annotated_frame, all_detections

    def should_skip_frame(self, start_time):
        """判断是否应该跳过当前帧"""
        processing_time = time.perf_counter() - start_time

        # 基于处理时间判断
        if processing_time > self.target_latency:
            return True

        # 基于FPS判断
        min_fps = 15
        if self.fps < min_fps:
            return True

        return False

    def handle_upload(self, annotated_frame, all_detections, current_time):
        """处理上传逻辑"""
        # 合并所有检测结果
        all_detections_list = []
        for model_detections in all_detections.values():
            all_detections_list.extend(model_detections)

        if len(all_detections_list) > 0 and (current_time - self.last_upload_time >= self.upload_interval):
            try:
                timestamp = int(current_time * 1000)
                filename = f"DJI_{timestamp}.jpg"
                self.upload_queue.put({
                    "image": annotated_frame.copy(),
                    "filename": filename,
                    "detection_data": all_detections_list,
                    "timestamp": current_time
                }, block=False)
                self.last_upload_time = current_time
            except queue.Full:
                logger.warning("上传队列已满，跳过上传")
            except Exception as e:
                logger.error(f"添加上传任务失败: {e}")

    def _upload_worker(self):
        """独立的图片上传工作线程"""
        logger.info("上传工作线程启动")
        output_dir = "output_frames"
        os.makedirs(output_dir, exist_ok=True)

        while self.upload_active or not self.upload_queue.empty():
            try:
                task = self.upload_queue.get(timeout=1.0)
                if task is None:
                    break

                start_time = time.time()
                image = task["image"]
                filename = task["filename"]
                detection_data = task["detection_data"]
                filepath = os.path.join(output_dir, filename)

                # 优化图片保存质量
                cv2.imwrite(filepath, image, [cv2.IMWRITE_JPEG_QUALITY, 85])

                try:
                    foldername = self.config['task']['taskname']
                    object_path = f"{foldername}/{filename}"
                    self.minio_uploader.upload_file(filepath, object_path)

                    payload = {
                        "taskid": self.taskid,
                        "path": object_path,
                        "tag": detection_data,
                        "aiid": self.aiid,
                    }

                    # 添加MQTT数据
                    if self.mqtt_enabled and self.mqtt_connected:
                        with self.mqtt_data_lock:
                            if self.latest_drone_data:
                                payload["drone_info"] = self.latest_drone_data

                    headers = {"Content-Type": "application/json"}
                    # 设置较短的超时时间
                    response = requests.post(self.res_api, json=payload, headers=headers, timeout=5)

                    if response.status_code == 200:
                        logger.debug(f"已上传帧至 MinIO: {object_path} | 耗时: {time.time() - start_time:.2f}s")
                    else:
                        logger.warning(f"API调用失败: {response.status_code} - {response.text}")
                except requests.exceptions.Timeout:
                    logger.warning(f"API调用超时: {self.res_api}")
                except Exception as e:
                    logger.error(f"上传/API调用失败: {e}")
                finally:
                    try:
                        os.remove(filepath)
                    except:
                        pass

                # 标记任务完成
                self.upload_queue.task_done()

            except queue.Empty:
                continue
            except Exception as e:
                logger.error(f"上传任务处理异常: {e}")

        logger.info("上传工作线程已停止")

    def on_mqtt_connect(self, client, userdata, flags, rc):
        """MQTT连接回调"""
        if rc == 0:
            client.subscribe(self.mqtt_topic)
            self.mqtt_connected = True
            logger.debug("MQTT连接状态正常")
        else:
            logger.error(f"MQTT连接失败，错误码: {rc}")
            self.mqtt_connected = False

    def on_mqtt_message(self, client, userdata, msg):
        """MQTT消息回调"""
        try:
            drone_data = json.loads(msg.payload.decode())
            with self.mqtt_data_lock:
                self.latest_drone_data = drone_data
            logger.debug(f"收到MQTT消息: {drone_data}")
        except Exception as e:
            logger.error(f"解析MQTT消息失败: {str(e)}")

    def start_mqtt_client(self):
        """启动MQTT客户端"""
        if not self.mqtt_enabled:
            logger.info("MQTT功能未启用")
            return False
        try:
            logger.info("启动MQTT客户端...")
            self.mqtt_client = mqtt.Client(client_id=self.mqtt_config.get('client_id', 'yolo_detection'))
            self.mqtt_client.on_connect = self.on_mqtt_connect
            self.mqtt_client.on_message = self.on_mqtt_message

            if 'username' in self.mqtt_config and 'password' in self.mqtt_config:
                self.mqtt_client.username_pw_set(
                    self.mqtt_config['username'],
                    self.mqtt_config['password']
                )

            self.mqtt_client.connect(
                self.mqtt_config['broker'],
                self.mqtt_config.get('port', 1883),
                self.mqtt_config.get('keepalive', 60)
            )

            self.mqtt_client.loop_start()
            logger.info("MQTT客户端已启动")
            return True
        except Exception as e:
            logger.error(f"启动MQTT客户端失败: {str(e)}")
            return False

    def stop_mqtt_client(self):
        """停止MQTT客户端"""
        if self.mqtt_client:
            try:
                self.mqtt_client.loop_stop()
                self.mqtt_client.disconnect()
                logger.info("MQTT客户端已停止")
            except Exception as e:
                logger.error(f"停止MQTT客户端失败: {str(e)}")
            finally:
                self.mqtt_client = None
                self.mqtt_connected = False

    def handle_reconnect(self):
        """处理RTMP重连"""
        if self.stop_event.is_set() or not self.running:
            logger.info("收到停止信号，跳过重连")
            return

        self.reconnect_attempts += 1
        if self.reconnect_attempts >= self.max_reconnect_attempts:
            logger.error("达到最大重连次数")
            self.running = False
            return

        # 使用指数退避策略计算延迟时间
        delay = min(30, self.reconnect_attempts * self.reconnect_delay * 2)  # 最大延迟30秒
        logger.warning(f"流中断，{delay}秒后重连 (第{self.reconnect_attempts}/{self.max_reconnect_attempts}次重连)")

        # 释放视频资源
        if self.cap:
            try:
                self.cap.release()
                self.cap = None
            except:
                pass

        # 在延迟期间检查停止信号
        start_time = time.time()
        while time.time() - start_time < delay:
            if self.stop_event.is_set() or not self.running:
                logger.info("收到停止信号，取消重连")
                return
            time.sleep(0.5)  # 增加休眠时间，减少CPU使用

        # 重置流稳定性计数器
        self.consecutive_read_failures = 0
        self.stream_stable = True
        self.stream_recovery_attempts += 1

        # 重新连接
        try:
            if self.stop_event.is_set() or not self.running:
                logger.info("收到停止信号，跳过重新连接")
                return

            logger.info("尝试重新连接RTMP...")
            if not self.initialize_rtmp():
                raise IOError("RTMP重连失败")

            logger.info("RTMP重连成功")
            self.reconnect_attempts = 0  # 重置重连次数
            self.stream_recovery_attempts = 0  # 重置恢复尝试次数

        except Exception as e:
            logger.error(f"重连异常: {str(e)}")

    def send_to_websocket(self, all_detections):
        """发送检测结果到WebSocket"""
        try:
            now = datetime.datetime.now()
            time_str = now.strftime("%H:%M:%S")

            # 合并所有模型的检测结果
            combined_detections = []
            for model_name, detections in all_detections.items():
                combined_detections.extend(detections)

            # 简化数据
            simplified_data = []
            for d in combined_detections:
                simplified_data.append({
                    'model_id': d['model_id'],
                    'model_name': d['model_name'],
                    'class_id': d['class_id'],
                    'class_name': d['class_name'],
                    'confidence': round(d['confidence'], 2),
                    'box': [round(float(c), 1) for c in d['box']],
                    'color': d['color']
                })

            # 添加流稳定性信息
            stream_info = {
                'stable': self.stream_stable,
                'consecutive_failures': self.consecutive_read_failures,
                'total_failures': self.total_read_failures,
                'recovery_attempts': self.stream_recovery_attempts
            }

            self.config['socketIO'].emit('detection_results', {
                'task_id': getattr(self, 'task_id', 'unknown'),
                'detections': simplified_data,
                'timestamp': time.time_ns() // 1000000,
                'fps': round(self.last_fps, 1),
                'frame_count': self.frame_count,
                'taskname': self.taskname,
                'time_str': time_str,
                'models_count': len(self.models),
                'stream_info': stream_info
            })
        except Exception as e:
            logger.error(f"WebSocket发送错误: {str(e)}")

    def check_stream_stability(self):
        """检查流稳定性"""
        current_time = time.time()
        if current_time - self.last_stream_check >= self.stream_check_interval:
            if self.consecutive_read_failures > 0:
                logger.warning(f"流稳定性警告: 连续读取失败 {self.consecutive_read_failures} 次")
                self.stream_stable = False
            else:
                self.stream_stable = True

            self.last_stream_check = current_time
            return True
        return False

    def handle_frame_read_failure(self):
        """处理帧读取失败"""
        self.consecutive_read_failures += 1
        self.total_read_failures += 1

        # 检查流稳定性
        self.check_stream_stability()

        # 根据连续失败次数采取不同策略
        if self.consecutive_read_failures == 1:
            # 第一次失败，短暂等待后重试
            logger.debug("帧读取失败，等待0.5秒后重试")
            time.sleep(self.retry_delay)
            return False  # 不立即重连

        elif self.consecutive_read_failures <= 3:
            # 2-3次失败，中等等待后重试
            logger.warning(f"连续 {self.consecutive_read_failures} 次帧读取失败，等待1秒后重试")
            time.sleep(1)
            return False  # 不立即重连

        elif self.consecutive_read_failures <= self.max_consecutive_failures:
            # 4-5次失败，较长时间等待后重试
            logger.error(f"连续 {self.consecutive_read_failures} 次帧读取失败，等待2秒后重试")
            time.sleep(2)
            return False  # 不立即重连

        else:
            # 超过最大失败次数，需要重连
            logger.error(f"连续 {self.consecutive_read_failures} 次帧读取失败，触发重连")
            return True  # 需要重连

    def run(self):
        """检测线程主循环"""
        try:
            logger.info(f"启动检测线程，任务ID: {getattr(self, 'task_id', 'unknown')}")
            # Windows系统检查
            if self._is_windows() and hasattr(self, 'windows_config'):
                if self.windows_config and 'error' in self.windows_config:
                    logger.warning(f"Windows配置问题: {self.windows_config['error']}")
            # 1. 加载多个模型
            if not self.load_models():
                logger.error("加密模型加载失败，线程退出")
                self.update_task_status('failed')
                return
            key_summary = self.get_key_verification_summary()
            logger.info(
                f"加密密钥验证结果: {key_summary['loaded_models']}/{key_summary['total_models']} 个模型加载成功")

            # 2. 初始化RTMP连接
            if not self.initialize_rtmp():
                logger.error("RTMP连接失败，线程退出")
                self.update_task_status('failed')
                return

            # 3. 初始化任务推流器
            if self.enable_push:
                if not self.initialize_task_streamer():
                    logger.warning(f"任务 {self.task_id} 推流器初始化失败，继续运行但不推流")
                    self.enable_push = False
            # 4. 启动推流管理器健康监控
            if self.enable_push and hasattr(self.stream_manager, 'start_health_monitor'):
                try:
                    self.stream_manager.start_health_monitor()
                    logger.info("推流健康监控已启动")
                except Exception as e:
                    logger.warning(f"启动健康监控失败: {e}")

            # 5. 启动MQTT（如果启用）
            if self.mqtt_enabled:
                self.start_mqtt_client()

            # 6. 启动上传线程
            self.upload_active = True
            self.upload_thread = threading.Thread(target=self._upload_worker, daemon=True)
            self.upload_thread.start()
            logger.info("图片上传线程已启动")

            # 7. 预热所有模型
            self.warmup_models()

            # 8. 更新任务状态
            self.initialized = True
            self.update_task_status('running')
            logger.info("资源初始化完成")

            # 9. 主循环
            while self.running and not self.stop_event.is_set():
                # 检查停止标志
                if self._force_stop or self._should_stop.is_set():
                    logger.info("收到停止信号，退出循环")
                    break

                if self.stop_event.is_set():
                    logger.info("收到停止事件信号，退出循环")
                    break

                start_time = time.perf_counter()

                try:
                    # 读取帧
                    ret, frame = self.cap.read()
                    if not ret:
                        if self.stop_event.is_set() or not self.running or self._force_stop:
                            logger.info("收到停止信号，不再尝试重连")
                            break

                        # 处理帧读取失败（带有重试和延迟策略）
                        need_reconnect = self.handle_frame_read_failure()

                        if need_reconnect:
                            # 需要重连
                            self.handle_reconnect()
                        continue

                    # 成功读取帧，重置失败计数器
                    if self.consecutive_read_failures > 0:
                        logger.info(f"恢复帧读取成功，之前连续失败 {self.consecutive_read_failures} 次")
                        self.consecutive_read_failures = 0
                        self.stream_stable = True

                    # 确保帧分辨率一致
                    if frame.shape[1] != self.original_width or frame.shape[0] != self.original_height:
                        frame = cv2.resize(frame, (self.original_width, self.original_height))

                    # 计算FPS
                    current_time = time.time()
                    time_diff = current_time - self.last_frame_time
                    if time_diff > 0:
                        self.fps = 0.9 * self.fps + 0.1 / time_diff
                        if current_time - self.last_status_update > 0.5:
                            self.last_fps = self.fps
                            self.last_status_update = current_time
                    self.last_frame_time = current_time

                    # 动态跳帧（Windows上更保守）
                    if self.should_skip_frame(start_time):
                        self.frame_count += 1
                        continue

                    # 多模型推理
                    annotated_frame, all_detections = self.multi_model_inference(frame)
                    self.detections_count = sum(len(dets) for dets in all_detections.values())

                    # 推流处理（Windows优化）
                    if self.enable_push:
                        if not self.push_frame_to_task_streamer(annotated_frame):
                            # 推流失败，但继续处理，避免影响检测
                            pass

                    # 上传处理
                    self.handle_upload(annotated_frame, all_detections, current_time)

                    # WebSocket发送
                    if current_time - self.last_log_time >= 1:
                        self.send_to_websocket(all_detections)
                        self.last_log_time = current_time

                    self.frame_count += 1
                    self.reconnect_attempts = 0

                    # 性能监控
                    elapsed = time.perf_counter() - start_time
                    self.processing_times.append(elapsed)
                    self.avg_process_time = self.avg_process_time * 0.9 + elapsed * 0.1

                    # 每50帧输出一次性能报告（包含推流统计）
                    if self.frame_count % 50 == 0:
                        avg_time = sum(self.processing_times) / len(
                            self.processing_times) if self.processing_times else 0
                        logger.info(
                            f"帧处理耗时: {avg_time * 1000:.2f}ms | 平均FPS: {(1.0 / avg_time if avg_time > 0 else 0):.2f}")

                        # 流稳定性报告
                        if self.total_read_failures > 0:
                            logger.info(
                                f"流稳定性: 总失败次数 {self.total_read_failures}, 当前连续失败 {self.consecutive_read_failures}")

                        # 推流统计
                        if self.enable_push:
                            logger.info(f"推流统计: {self.stream_stats['total_frames_pushed']}帧, "
                                        f"成功率: {self.stream_stats['push_success_rate']:.2%}, "
                                        f"重启: {self.stream_stats.get('ffmpeg_restarts', 0)}次")
                        self.processing_times = []


                except Exception as e:
                    logger.error(f"处理帧时发生异常: {str(e)}")
                    if self.stop_event.is_set() or not self.running or self._force_stop:
                        logger.info("收到停止信号，退出异常处理")
                        break
                    continue

            logger.info("检测线程主循环结束")

        except Exception as e:
            logger.error(f"检测线程异常: {str(e)}")
            logger.error(traceback.format_exc())
        finally:
            self.cleanup()
            logger.info("检测线程已安全停止")
            self.update_task_status('stopped')
            return False

    def stop(self):
        """停止检测线程"""
        logger.info(f"收到停止请求，任务ID: {getattr(self, 'task_id', 'unknown')}")

        # 首先设置强制停止标志
        self._force_stop = True
        self._should_stop.set()  # 设置停止事件

        # 发送停止事件
        self.stop_event.set()

        # 设置运行标志为False
        self.running = False

        # 强制释放视频流以解除阻塞
        if self.cap:
            try:
                # 强制中断视频读取
                self.cap.release()
                logger.info("已强制释放视频流")
            except Exception as e:
                logger.error(f"释放视频流失败: {str(e)}")

        # 清理任务推流器
        self.cleanup_task_streamer()

        # 停止MQTT客户端
        if self.mqtt_enabled:
            self.stop_mqtt_client()

        # 停止上传线程
        self.upload_active = False
        if hasattr(self, 'upload_thread') and self.upload_thread and self.upload_thread.is_alive():
            # 发送停止信号到上传队列
            try:
                self.upload_queue.put(None, timeout=0.5)
            except:
                pass

        logger.info(f"停止信号已发送，任务ID: {getattr(self, 'task_id', 'unknown')}")

    def cleanup(self):
        """清理所有资源"""
        logger.info(f"开始清理资源，任务ID: {getattr(self, 'task_id', 'unknown')}")

        if hasattr(self, '_cleaning_up') and self._cleaning_up:
            logger.warning("资源已经在清理中，跳过重复清理")
            return

        self._cleaning_up = True

        try:
            # 停止MQTT客户端
            if self.mqtt_enabled:
                self.stop_mqtt_client()

            # 停止上传线程
            self.upload_active = False
            if hasattr(self, 'upload_thread') and self.upload_thread and self.upload_thread.is_alive():
                logger.info("停止上传线程...")
                try:
                    if hasattr(self, 'upload_queue'):
                        try:
                            self.upload_queue.put(None, timeout=0.5)
                        except queue.Full:
                            try:
                                while not self.upload_queue.empty():
                                    self.upload_queue.get_nowait()
                            except:
                                pass
                            self.upload_queue.put(None, block=False)
                        except:
                            pass

                    self.upload_thread.join(3.0)
                    if self.upload_thread.is_alive():
                        logger.warning("上传线程未在3秒内停止")
                except Exception as e:
                    logger.error(f"停止上传线程异常: {str(e)}")
                finally:
                    self.upload_thread = None

            # 清理任务推流器
            self.cleanup_task_streamer()

            # 释放视频流
            if hasattr(self, 'cap') and self.cap:
                logger.info("释放视频流...")
                try:
                    self.cap.release()
                except Exception as e:
                    logger.error(f"释放视频流异常: {str(e)}")
                finally:
                    self.cap = None

            # 释放所有模型
            logger.info(f"释放所有模型，共 {len(self.models)} 个")
            for i, model_info in enumerate(self.models):
                try:
                    model = model_info['model']
                    model_name = model_info['name']

                    # 清理模型缓存
                    if hasattr(model, 'predictor'):
                        try:
                            del model.predictor
                        except:
                            pass
                    if hasattr(model, 'model'):
                        try:
                            del model.model
                        except:
                            pass

                    # 释放模型引用
                    del model
                    logger.info(f"模型 {model_name} 已释放")
                except Exception as e:
                    logger.error(f"释放模型 {i} 异常: {str(e)}")

            self.models = []

            # 清理GPU缓存
            logger.info("清理GPU缓存...")
            try:
                if torch.cuda.is_available():
                    torch.cuda.empty_cache()
                    gc.collect()
                    logger.info("GPU缓存已清理")
            except Exception as e:
                logger.error(f"清理GPU缓存异常: {str(e)}")

            # 清理其他资源
            logger.info("清理其他资源...")
            try:
                if hasattr(self, 'upload_queue'):
                    try:
                        while not self.upload_queue.empty():
                            try:
                                self.upload_queue.get_nowait()
                            except:
                                break
                    except:
                        pass

                self.processing_times = []

                logger.info("其他资源已清理")
            except Exception as e:
                logger.error(f"清理其他资源异常: {str(e)}")

            logger.info("资源清理完成")

        except Exception as e:
            logger.error(f"资源清理过程中发生异常: {str(e)}")
            logger.error(traceback.format_exc())
        finally:
            self._cleaning_up = False  # 重置清理标志

    def update_task_status(self, status):
        """更新任务状态"""
        if hasattr(self, 'task_id') and self.task_id:
            try:
                from task_manager import task_manager
                if hasattr(task_manager, 'update_task_status'):
                    task_manager.update_task_status(self.task_id, status)
                else:
                    # 备用方法：直接更新全局数据
                    tasks_dict = gd.get_or_create_dict('tasks')
                    if self.task_id in tasks_dict:
                        tasks_dict[self.task_id]['status'] = status
                        logger.debug(f"更新任务 {self.task_id} 状态为: {status}")
            except Exception as e:
                logger.warning(f"更新任务状态失败: {str(e)}")