yolov8_live_rtmp_stream_detection.py

# yolov8_live_rtmp_stream_detection.py
# ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
# https://github.com/FlyingFathead/dvr-yolov8-detection
# ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

# get the version number
import version  # Import the version module
version_number = version.version_number

import av
import cv2
import torch
import logging
import numpy as np
from utils import hz_line

# Time and timezone related
import time
from datetime import datetime
import pytz

from collections import deque, defaultdict
import pyttsx3
import os
from ultralytics import YOLO
import threading
from threading import Thread, Event, Lock
from queue import Queue, Empty, Full
import configparser
import argparse
import signal
import sys

# Import web server functions
from web_server import start_web_server, set_output_frame

# Import the remote sync module
import remote_sync

# Import sending alerts on Telegram
import telegram_alerts

hz_line()
print(f"::: dvr-yolov8-detection v{version_number} | https://github.com/FlyingFathead/dvr-yolov8-detection/")
hz_line()

# Shared data structures
detections_list = deque(maxlen=100)  # Store up to 100 latest detections on web UI
logs_list = deque(maxlen=100)        # Store up to 100 latest logs on web UI
detection_timestamps = deque(maxlen=10000)  # Store up to 10,000 timestamps

# Global variable to store masked regions
masked_regions = []  # each entry: { "name": str, "confidence_threshold": float, "x1": int, "y1": int, "x2": int, "y2": int }

# Locks for thread safety
timestamps_lock = threading.Lock()
detections_lock = threading.Lock()
logs_lock = threading.Lock()

# Load configuration from `config.ini` file with case-sensitive keys
def load_config(config_path=None):
    try:
        if config_path is None:
            script_dir = os.path.dirname(os.path.abspath(__file__))
            print(f"Script directory resolved to: {script_dir}", flush=True)
            config_path = os.path.join(script_dir, 'config.ini')
        
        print(f"Attempting to load config from: {config_path}", flush=True)
        config = configparser.ConfigParser()
        # config.optionxform = str  # Make keys case-sensitive
        read_files = config.read(config_path)
        
        if not read_files:
            print(f"Warning: Config file {config_path} not found or is empty. Using default settings.", flush=True)
        else:
            print(f"Config file {config_path} loaded successfully.", flush=True)
        
        return config
    except Exception as e:
        print(f"Error loading config: {e}", flush=True)
        raise

# Shared data structure to hold image filenames associated with detection counts
detection_images = defaultdict(list)
detection_images_lock = threading.Lock()

# Load configuration
print("Starting configuration loading...", flush=True)
config = load_config()
print("Configuration loading completed.", flush=True)

# Assign configurations to variables
HEADLESS = config.getboolean('detection', 'headless', fallback=False)
DEFAULT_CONF_THRESHOLD = config.getfloat('general', 'default_conf_threshold')
DEFAULT_MODEL_VARIANT = config.get('general', 'default_model_variant')
USE_WEBCAM = config.getboolean('input', 'use_webcam', fallback=False)
WEBCAM_INDEX = config.getint('input', 'webcam_index', fallback=0)
STREAM_URL = config.get('stream', 'stream_url')
DRAW_RECTANGLES = config.getboolean('detection', 'draw_rectangles')
SAVE_FULL_FRAMES = config.getboolean('detection', 'save_full_frames', fallback=True)
FULL_FRAME_IMAGE_FORMAT = config.get('detection', 'full_frame_image_format', fallback='jpg')
DETECTION_AREA_IMAGE_FORMAT = config.get('detection', 'detection_area_image_format', fallback='jpg')
IMAGE_QUALITY = config.getint('detection', 'image_quality', fallback=100)
PNG_COMPRESSION_LEVEL = config.getint('detection', 'png_compression_level', fallback=3)
WEBP_LOSSLESS = config.getboolean('detection', 'webp_lossless', fallback=False)
SAVE_DETECTION_AREAS = config.getboolean('detection', 'save_detection_areas', fallback=False)
DETECTION_AREA_MARGIN = config.getint('detection', 'detection_area_margin', fallback=0)
IMAGE_SAVE_QUEUE_MAXSIZE = config.getint('performance', 'IMAGE_SAVE_QUEUE_MAXSIZE', fallback=1000)
USE_ENV_SAVE_DIR = config.getboolean('detection', 'use_env_save_dir', fallback=False)
ENV_SAVE_DIR_VAR = config.get('detection', 'env_save_dir_var', fallback='YOLO_SAVE_DIR')
# Extract SAVE_DIR_BASE from config.ini or set a default value
SAVE_DIR_BASE = config.get('detection', 'save_dir_base', fallback='./yolo_detections')
DEFAULT_SAVE_DIR = config.get('detection', 'default_save_dir')
FALLBACK_SAVE_DIR = config.get('detection', 'fallback_save_dir')
RETRY_DELAY = config.getint('detection', 'retry_delay')
MAX_RETRIES = config.getint('detection', 'max_retries')
RESCALE_INPUT = config.getboolean('detection', 'rescale_input')
TARGET_HEIGHT = config.getint('detection', 'target_height')
DENOISE = config.getboolean('detection', 'denoise')
USE_PROCESS_FPS = config.getboolean('detection', 'use_process_fps')
PROCESS_FPS = config.getint('detection', 'process_fps')
TIMEOUT = config.getint('detection', 'timeout')
TTS_COOLDOWN = config.getint('detection', 'tts_cooldown')
# Logging options
ENABLE_DETECTION_LOGGING_TO_FILE = config.getboolean('logging', 'enable_detection_logging_to_file')
LOG_DIRECTORY = config.get('logging', 'log_directory')
LOG_FILE = config.get('logging', 'log_file')
DETECTION_LOG_FILE = config.get('logging', 'detection_log_file')
# New configuration for creating date-based subdirectories
CREATE_DATE_SUBDIRS = config.getboolean('detection', 'create_date_subdirs', fallback=False)
# Web server configuration
ENABLE_WEBSERVER = config.getboolean('webserver', 'enable_webserver', fallback=False)
WEBSERVER_HOST = config.get('webserver', 'webserver_host', fallback='0.0.0.0')
WEBSERVER_PORT = config.getint('webserver', 'webserver_port', fallback=5000)

# Read AGGREGATED_DETECTIONS_FILE from config for remote sync
ENABLE_PERSISTENT_AGGREGATED_DETECTIONS = config.getboolean('aggregation', 'enable_persistent_aggregated_detections', fallback=False)
AGGREGATED_DETECTIONS_FILE = config.get('aggregation', 'aggregated_detections_file', fallback='./logs/aggregated_detections.json')
# max number of aggregation entries to show on webUI
webui_max_aggregation_entries = 100

# Initialize the image save queue and stop event
# image_save_queue = Queue()
image_save_queue = Queue(maxsize=IMAGE_SAVE_QUEUE_MAXSIZE)
image_saving_stop_event = threading.Event()

# load masked regions if enabled
def load_masked_regions(config, main_logger):
    """
    If 'enable_masked_regions' is True, attempt to load the JSON
    from 'output_json'. Store in global 'masked_regions'.
    """
    global masked_regions

    enable_masked_regions = config.getboolean('region_masker', 'enable_masked_regions', fallback=False)
    if not enable_masked_regions:
        main_logger.info("Masked regions are disabled in config. Proceeding without any region masking.")
        return

    # If enabled, read the JSON file path
    masked_json_path = config.get('region_masker', 'output_json', fallback='./ignore_zones.json')

    main_logger.info(f"Masked regions are ENABLED. Attempting to load from: {masked_json_path}")

    if not os.path.exists(masked_json_path):
        main_logger.warning(f"Masked regions file '{masked_json_path}' not found. No masked regions will be used.")
        return

    try:
        import json
        with open(masked_json_path, 'r') as f:
            data = json.load(f)
        loaded_zones = data.get("ignore_zones", [])
        masked_regions = loaded_zones  # store globally
        if loaded_zones:
            main_logger.info(f"Loaded {len(loaded_zones)} masked region(s) from '{masked_json_path}':")
            for i, zone in enumerate(loaded_zones, 1):
                zname = zone.get("name", f"Zone_{i}")
                cthr = zone.get("confidence_threshold", 0.0)
                x1 = zone.get("x1", 0)
                y1 = zone.get("y1", 0)
                x2 = zone.get("x2", 0)
                y2 = zone.get("y2", 0)
                main_logger.info(f"  -> {zname} (threshold={cthr}), coords=({x1},{y1}) to ({x2},{y2})")
        else:
            main_logger.info(f"File '{masked_json_path}' loaded but no zones found. Proceeding with no masked regions.")
    except Exception as e:
        main_logger.error(f"Failed to parse masked regions JSON '{masked_json_path}': {e}")
        main_logger.info("Continuing with no masked regions loaded.")

# masked regions
def apply_masked_regions(x1, y1, x2, y2, confidence, masked_regions):
    """
    Example logic: if detection box intersects a masked region,
    then either ignore it or require a higher confidence threshold.
    This is TOTALLY up to you how you interpret it.

    Returns (keep_detection_bool, skip_info_dict|None).
    skip_info_dict has keys like {"zone_name": str, "required_conf": float} if skipping.
    """
    for zone in masked_regions:
        zx1 = zone.get("x1", 0)
        zy1 = zone.get("y1", 0)
        zx2 = zone.get("x2", 0)
        zy2 = zone.get("y2", 0)
        if boxes_intersect(x1, y1, x2, y2, zx1, zy1, zx2, zy2):
            zone_name = zone.get("name", "UnknownZone")
            required_conf = zone.get("confidence_threshold", 1.0)
            # If detection's conf < the zone's threshold => skip
            if confidence < required_conf:
                return (False, {
                    "zone_name": zone_name,
                    "required_conf": required_conf
                })
    return (True, None)

    # for zone in masked_regions:
    #     # Parse region coords
    #     zx1 = zone.get("x1", 0)
    #     zy1 = zone.get("y1", 0)
    #     zx2 = zone.get("x2", 0)
    #     zy2 = zone.get("y2", 0)
    #     # Could do fancy polygon intersection, but let's do simple AABB
    #     if boxes_intersect(x1, y1, x2, y2, zx1, zy1, zx2, zy2):
    #         # Let’s say zone.confidence_threshold acts as "ignore unless conf >= that threshold"
    #         required_conf = zone.get("confidence_threshold", 1.0)  
    #         if confidence < required_conf:
    #             return False  # means "skip this detection"
    # return True

def boxes_intersect(ax1, ay1, ax2, ay2, bx1, by1, bx2, by2):
    # quick axis-aligned overlap check
    # assume (x1 < x2) and (y1 < y2)
    if ax2 < bx1 or bx2 < ax1:
        return False  # no horizontal overlap
    if ay2 < by1 or by2 < ay1:
        return False  # no vertical overlap
    return True

# List handler for webUI logging
class ListHandler(logging.Handler):
    def __init__(self, logs_list, logs_lock):
        super().__init__()
        self.logs_list = logs_list
        self.logs_lock = logs_lock

    def emit(self, record):
        log_entry = self.format(record)
        with self.logs_lock:
            self.logs_list.append(log_entry)

# Centralized Logging Configuration
def setup_logging():
    # Create a root logger with StreamHandler
    root_logger = logging.getLogger()
    root_logger.setLevel(logging.INFO)
    
    # Prevent root logger from propagating to higher loggers (if any)
    root_logger.propagate = False
    
    # Define and configure the logger for remote_sync
    remote_sync_logger = logging.getLogger("remote_sync")
    remote_sync_logger.setLevel(logging.INFO)
    remote_sync_logger.propagate = False

    # Create formatter
    formatter = logging.Formatter('%(asctime)s - %(levelname)s - %(name)s - %(message)s')
    
    # Create and add StreamHandler to root logger
    stream_handler = logging.StreamHandler(sys.stdout)
    stream_handler.setLevel(logging.INFO)
    stream_handler.setFormatter(formatter)
    root_logger.addHandler(stream_handler)
    
    # Define the main logger
    main_logger = logging.getLogger('main')
    main_logger.setLevel(logging.INFO)
    main_logger.propagate = False  # Prevent propagation to root
    
    # Add ListHandler to main_logger to capture WARNING and ERROR logs
    list_handler = ListHandler(logs_list, logs_lock)
    list_handler.setLevel(logging.WARNING)  # Capture WARNING and ERROR logs
    list_handler.setFormatter(formatter)
    main_logger.addHandler(list_handler)
    
    # **Add StreamHandler to main_logger for INFO level logs**
    stream_handler_main = logging.StreamHandler(sys.stdout)
    stream_handler_main.setLevel(logging.INFO)
    stream_handler_main.setFormatter(formatter)
    main_logger.addHandler(stream_handler_main)
    
    # Define detection logger
    detection_logger = logging.getLogger('detection')
    detection_logger.setLevel(logging.INFO)
    detection_logger.propagate = False  # Prevent propagation to root
    
    # **Configure 'remote_sync' logger to propagate to root logger**
    remote_sync_logger = logging.getLogger("remote_sync")
    remote_sync_logger.setLevel(logging.INFO)
    remote_sync_logger.propagate = True  # Allow propagation to root logger

    # Initialize detection_log_path
    detection_log_path = None
    
    # Add FileHandler to detection_logger if enabled
    if ENABLE_DETECTION_LOGGING_TO_FILE:
        if not os.path.exists(LOG_DIRECTORY):
            os.makedirs(LOG_DIRECTORY)
        detection_log_path = os.path.join(LOG_DIRECTORY, DETECTION_LOG_FILE)
        detection_file_handler = logging.FileHandler(detection_log_path)
        detection_file_handler.setLevel(logging.INFO)
        detection_file_handler.setFormatter(formatter)
        detection_logger.addHandler(detection_file_handler)
        main_logger.info(f"Detection logging to file is enabled. Logging to: {detection_log_path}")
    else:
        main_logger.info("Detection logging to file is disabled.")
    
    # Define web_server logger
    web_server_logger = logging.getLogger('web_server')
    web_server_logger.setLevel(logging.INFO)
    web_server_logger.propagate = False  # Prevent propagation to root
    
    # Add StreamHandler to web_server_logger if not already present
    if not web_server_logger.hasHandlers():
        web_stream_handler = logging.StreamHandler(sys.stdout)
        web_stream_handler.setLevel(logging.INFO)
        web_stream_handler.setFormatter(formatter)
        web_server_logger.addHandler(web_stream_handler)
    
    return main_logger, detection_logger, web_server_logger, detection_log_path

# Initialize logging
main_logger, detection_logger, web_server_logger, detection_log_path = setup_logging()

# Load masked regions if enabled
load_masked_regions(config, main_logger)

# Timekeeping and frame counting
last_log_time = time.time()

# Initialize TTS engine
tts_engine = pyttsx3.init()
tts_lock = Lock()
last_tts_time = 0
tts_thread = None
tts_stop_event = Event()

# Check if Telegram alerts are enabled
if telegram_alerts.bot is None:
    main_logger.warning("Telegram alerts are disabled.")
else:
    main_logger.info("Telegram alerts are enabled.")

# Initialize remote sync (if enabled)
remote_sync_obj = remote_sync.RemoteSync(config, main_logger, SAVE_DIR_BASE, AGGREGATED_DETECTIONS_FILE)
remote_sync_obj.start()

# Specify supported image formats
SUPPORTED_IMAGE_FORMATS = ['jpg', 'jpeg', 'png', 'webp']

# Load the YOLOv8 model
def load_model(model_variant=DEFAULT_MODEL_VARIANT):
    """Loads the YOLO model and puts it onto the configured CUDA device (or CPU)."""
    try:
        model = YOLO(model_variant)

        # 1) Read from config
        cuda_device_id = config.getint('hardware', 'cuda_device_id', fallback=0)
        cuda_fallback_if_unavailable = config.getboolean('hardware', 'cuda_fallback_if_unavailable', fallback=True)

        if torch.cuda.is_available():
            # 2) Check how many devices we have
            num_gpus = torch.cuda.device_count()

            if cuda_device_id < num_gpus:
                try:
                    # Attempt to put the model on the user-specified device
                    device_str = f"cuda:{cuda_device_id}"
                    model.to(device_str)
                    main_logger.info(f"Using CUDA device {cuda_device_id} for model inference.")
                except Exception as e:
                    main_logger.error(f"Failed to load model on cuda:{cuda_device_id}: {e}")
                    if cuda_fallback_if_unavailable and num_gpus > 0:
                        # Attempt fallback on the first available GPU (usually cuda:0)
                        main_logger.warning("Falling back to cuda:0")
                        model.to("cuda:0")
                    else:
                        main_logger.warning("No fallback GPU or fallback disabled. Using CPU instead.")
                        model.to('cpu')
            else:
                # The user-specified device doesn't exist
                main_logger.warning(f"Requested cuda:{cuda_device_id}, but only {num_gpus} GPU(s) found.")
                if cuda_fallback_if_unavailable and num_gpus > 0:
                    main_logger.warning("Falling back to cuda:0")
                    model.to("cuda:0")
                else:
                    main_logger.warning("No fallback GPU or fallback disabled. Using CPU instead.")
                    model.to('cpu')
        else:
            # No GPU present at all
            main_logger.warning("CUDA not available, using CPU for model inference.")
            model.to('cpu')

        return model
    except Exception as e:
        logging.error(f"Error loading model {model_variant}: {e}")
        raise

# Initialize model
model = load_model(DEFAULT_MODEL_VARIANT)

# Get available CUDA GPU and its details
def log_cuda_info():
    if not torch.cuda.is_available():
        logging.warning("No CUDA devices detected. Running on CPU.")
        return

    num_gpus = torch.cuda.device_count()
    logging.info(f"Number of CUDA devices detected: {num_gpus}")

    for i in range(num_gpus):
        gpu_name = torch.cuda.get_device_name(i)
        gpu_capability = torch.cuda.get_device_capability(i)
        gpu_memory = torch.cuda.get_device_properties(i).total_memory / 1e9  # Convert bytes to GB
        logging.info(f"CUDA Device {i}: {gpu_name}")
        logging.info(f"  Compute Capability: {gpu_capability[0]}.{gpu_capability[1]}")
        logging.info(f"  Total Memory: {gpu_memory:.2f} GB")

    # // not in use after 0.1614.3, preferred CUDA device is now set in `config.ini`
    # Log the current device being used 
    # current_device = torch.cuda.current_device()
    # current_gpu_name = torch.cuda.get_device_name(current_device)    
    # logging.info(f"Using CUDA Device {current_device}: {current_gpu_name}")

# Function to get the base save directory (without date subdirs)
def get_base_save_dir():
    base_save_dir = None

    # 1. Use environment-specified directory
    if USE_ENV_SAVE_DIR:
        env_dir = os.getenv(ENV_SAVE_DIR_VAR)
        if env_dir and os.path.exists(env_dir) and os.access(env_dir, os.W_OK):
            main_logger.info(f"Using environment-specified save directory: {env_dir}")
            base_save_dir = env_dir  # Set base_save_dir here
        else:
            main_logger.warning(
                f"Environment variable {ENV_SAVE_DIR_VAR} is set but the directory does not exist or is not writable. Checked path: {env_dir}"
            )

    # 2. Use save_dir_base from config or default
    if not base_save_dir:
        SAVE_DIR_BASE = config.get('detection', 'save_dir_base', fallback='./yolo_detections')
        if os.path.exists(SAVE_DIR_BASE) and os.access(SAVE_DIR_BASE, os.W_OK):
            main_logger.info(f"Using save_dir_base from config: {SAVE_DIR_BASE}")
            base_save_dir = SAVE_DIR_BASE
        else:
            main_logger.warning(f"save_dir_base {SAVE_DIR_BASE} does not exist or is not writable. Attempting to create it.")
            try:
                os.makedirs(SAVE_DIR_BASE, exist_ok=True)
                if os.access(SAVE_DIR_BASE, os.W_OK):
                    main_logger.info(f"Created and using save_dir_base: {SAVE_DIR_BASE}")
                    base_save_dir = SAVE_DIR_BASE
                else:
                    main_logger.warning(f"save_dir_base {SAVE_DIR_BASE} is not writable after creation.")
            except Exception as e:
                main_logger.error(f"Failed to create save_dir_base: {SAVE_DIR_BASE}. Error: {e}")

    # 3. Fallback to fallback_save_dir
    if not base_save_dir and FALLBACK_SAVE_DIR:
        if os.path.exists(FALLBACK_SAVE_DIR) and os.access(FALLBACK_SAVE_DIR, os.W_OK):
            main_logger.info(f"Using fallback save directory: {FALLBACK_SAVE_DIR}")
            base_save_dir = FALLBACK_SAVE_DIR
        else:
            main_logger.warning(f"Fallback save directory {FALLBACK_SAVE_DIR} does not exist or is not writable. Attempting to create it.")
            try:
                os.makedirs(FALLBACK_SAVE_DIR, exist_ok=True)
                if os.access(FALLBACK_SAVE_DIR, os.W_OK):
                    main_logger.info(f"Created and using fallback save directory: {FALLBACK_SAVE_DIR}")
                    base_save_dir = FALLBACK_SAVE_DIR
                else:
                    main_logger.warning(f"Fallback save directory {FALLBACK_SAVE_DIR} is not writable after creation.")
            except Exception as e:
                main_logger.error(f"Failed to create fallback save directory: {FALLBACK_SAVE_DIR}. Error: {e}")

    # 4. Use default_save_dir
    if not base_save_dir and DEFAULT_SAVE_DIR:
        if os.path.exists(DEFAULT_SAVE_DIR) and os.access(DEFAULT_SAVE_DIR, os.W_OK):
            main_logger.info(f"Using default save directory: {DEFAULT_SAVE_DIR}")
            base_save_dir = DEFAULT_SAVE_DIR
        else:
            main_logger.warning(f"Default save directory {DEFAULT_SAVE_DIR} does not exist or is not writable. Attempting to create it.")
            try:
                os.makedirs(DEFAULT_SAVE_DIR, exist_ok=True)
                if os.access(DEFAULT_SAVE_DIR, os.W_OK):
                    main_logger.info(f"Created and using default save directory: {DEFAULT_SAVE_DIR}")
                    base_save_dir = DEFAULT_SAVE_DIR
                else:
                    main_logger.warning(f"Default save directory {DEFAULT_SAVE_DIR} is not writable after creation.")
            except Exception as e:
                main_logger.error(f"Failed to create default save directory: {DEFAULT_SAVE_DIR}. Error: {e}")

    # 5. Final fallback to a hardcoded directory (optional)
    if not base_save_dir:
        final_fallback_dir = os.path.join(os.path.dirname(__file__), 'final_fallback_detections')
        if os.path.exists(final_fallback_dir) and os.access(final_fallback_dir, os.W_OK):
            main_logger.info(f"Using final hardcoded fallback save directory: {final_fallback_dir}")
            base_save_dir = final_fallback_dir
        else:
            try:
                os.makedirs(final_fallback_dir, exist_ok=True)
                if os.access(final_fallback_dir, os.W_OK):
                    main_logger.info(f"Created and using final hardcoded fallback save directory: {final_fallback_dir}")
                    base_save_dir = final_fallback_dir
                else:
                    main_logger.warning(f"Final fallback save directory {final_fallback_dir} is not writable after creation.")
            except Exception as e:
                main_logger.error(f"Failed to create final hardcoded fallback save directory: {final_fallback_dir}. Error: {e}")

    # Raise error if no writable directory is found
    if not base_save_dir:
        raise RuntimeError("No writable save directory available.")

    return base_save_dir

# Initialize SAVE_DIR and CURRENT_DATE
# Initialize SAVE_DIR_BASE using the updated get_base_save_dir function
SAVE_DIR_BASE = get_base_save_dir()
main_logger.info(f"SAVE_DIR_BASE is set to: {SAVE_DIR_BASE}")
CURRENT_DATE = datetime.now().date()

# Function to get the current save directory with date-based subdirectories
def get_current_save_dir():
    global CURRENT_DATE, SAVE_DIR_BASE

    if CREATE_DATE_SUBDIRS:
        new_date = datetime.now().date()
        if new_date != CURRENT_DATE:
            CURRENT_DATE = new_date
            # Update SAVE_DIR with new date-based subdirectories
            year = CURRENT_DATE.strftime("%Y")
            month = CURRENT_DATE.strftime("%m")
            day = CURRENT_DATE.strftime("%d")
            date_subdir = os.path.join(SAVE_DIR_BASE, year, month, day)
            try:
                os.makedirs(date_subdir, exist_ok=True)
                main_logger.info(f"Date changed. Created/navigated to new date-based subdirectory: {date_subdir}")
                return date_subdir
            except Exception as e:
                main_logger.error(f"Failed to create new date-based subdirectory {date_subdir}: {e}")
                # Fallback to base_save_dir if subdirectory creation fails
                return SAVE_DIR_BASE
        else:
            # Ensure the date-based subdirectory exists
            year = CURRENT_DATE.strftime("%Y")
            month = CURRENT_DATE.strftime("%m")
            day = CURRENT_DATE.strftime("%d")
            date_subdir = os.path.join(SAVE_DIR_BASE, year, month, day)
            if not os.path.exists(date_subdir):
                try:
                    os.makedirs(date_subdir, exist_ok=True)
                    main_logger.info(f"Created date-based subdirectory: {date_subdir}")
                except Exception as e:
                    main_logger.error(f"Failed to create date-based subdirectory {date_subdir}: {e}")
            return date_subdir
    else:
        return SAVE_DIR_BASE

# Initialize the initial SAVE_DIR
SAVE_DIR = get_current_save_dir()

# Function to log detection details
def log_detection_details(detections, frame_count, timestamp):
    if ENABLE_DETECTION_LOGGING_TO_FILE:
        for detection in detections:
            x1, y1, x2, y2, confidence, class_idx = detection
            detection_logger.info(f"Detection: Frame {frame_count}, Timestamp {timestamp}, Coordinates: ({x1}, {y1}), ({x2}, {y2}), Confidence: {confidence:.2f}")

# Function to resize while maintaining aspect ratio
def resize_frame(frame, target_height):
    """Resize the frame to the target height while maintaining the aspect ratio."""
    height, width = frame.shape[:2]
    aspect_ratio = width / height
    new_width = int(target_height * aspect_ratio)
    resized_frame = cv2.resize(frame, (new_width, target_height))
    return resized_frame

# Function to announce detection using a separate thread
def announce_detection():
    global last_tts_time
    while not tts_stop_event.is_set():
        current_time = time.time()
        if current_time - last_tts_time >= TTS_COOLDOWN:
            with tts_lock:
                tts_engine.say("Human detected!")
                tts_engine.runAndWait()
            last_tts_time = current_time
        time.sleep(1)

# # Function to save detection image with date check
# def save_detection_image(frame, detection_count):
#     global SAVE_DIR    
#     main_logger.info(f"Current SAVE_DIR is: {SAVE_DIR}")
#     main_logger.info("Attempting to save detection image.")    
#     try:
#         # Update SAVE_DIR based on current date
#         SAVE_DIR = get_current_save_dir()
        
#         timestamp = datetime.now().strftime("%Y%m%d-%H%M%S")  # Format timestamp as YYYYMMDD-HHMMSS
#         filename = os.path.join(SAVE_DIR, f"{timestamp}_{detection_count}.{IMAGE_FORMAT}")  # Ensure SAVE_DIR already includes date subdirs
#         if not cv2.imwrite(filename, frame):
#             main_logger.error(f"Failed to save detection image: {filename}")
#         else:
#             main_logger.info(f"Saved detection image: {filename}")
#     except Exception as e:
#         main_logger.error(f"Error saving detection image: {e}")

# Check if the CUDA denoising function is available
def cuda_denoising_available():
    try:
        _ = cv2.cuda.createFastNlMeansDenoisingColored
        main_logger.info("CUDA denoising is available.")
        return True
    except AttributeError:
        main_logger.warning("CUDA denoising is not available.")
        return False

# Frame capture thread
def frame_capture_thread(stream_url, use_webcam, webcam_index, frame_queue, stop_event):
    retries = 0
    cap = None

    if use_webcam:
        main_logger.info(f"Using webcam with index {webcam_index}")
        cap = cv2.VideoCapture(webcam_index)
        if not cap.isOpened():
            main_logger.error(f"Unable to open webcam with index {webcam_index}")
            return
        while not stop_event.is_set():
            ret, frame = cap.read()
            if not ret:
                main_logger.warning("Failed to read frame from webcam.")
                time.sleep(1)
                continue
            frame_queue.put(frame)
        cap.release()
        main_logger.info("Webcam video source closed.")
    else:
        while retries < MAX_RETRIES and not stop_event.is_set():
            try:
                main_logger.info(f"Attempting to open video stream: {stream_url}")
                container = av.open(stream_url)
                main_logger.info("Successfully opened video stream.")

                for frame in container.decode(video=0):
                    if stop_event.is_set():
                        break
                    image = frame.to_ndarray(format='bgr24')
                    frame_queue.put(image)
                # If the stream ends, break out of the loop
                break
            except av.AVError as e:
                main_logger.error(f"Error reading video stream: {e}")
                retries += 1
                time.sleep(RETRY_DELAY)
                main_logger.info(f"Retrying... ({retries}/{MAX_RETRIES})")
            except Exception as e:
                main_logger.error(f"Unexpected error: {e}")
                retries += 1
                time.sleep(RETRY_DELAY)
                main_logger.info(f"Retrying... ({retries}/{MAX_RETRIES})")
        if retries >= MAX_RETRIES:
            main_logger.error(f"Failed to open video source after {MAX_RETRIES} retries.")
            return
        main_logger.info("RTMP video source closed.")

# Frame processing thread
def frame_processing_thread(
    frame_queue, stop_event,
    conf_threshold, draw_rectangles, denoise,
    process_fps, use_process_fps,
    detection_ongoing, headless
):
    global last_tts_time, tts_thread, tts_stop_event

    model = load_model(DEFAULT_MODEL_VARIANT)
    use_cuda_denoising = cuda_denoising_available()
    detection_count = 0
    last_log_time = time.time()
    total_frames = 0
    detecting_human = False

    if not headless:
        cv2.namedWindow('Real-time Human Detection', cv2.WINDOW_NORMAL)

    try:
        while not stop_event.is_set() or not frame_queue.empty():
            if frame_queue.empty():
                # If queue has no frames, just sleep briefly
                time.sleep(0.01)
                continue

            frame = frame_queue.get()
            start_time = time.time()

            # 1) Optional resize
            if RESCALE_INPUT:
                resized_frame = resize_frame(frame, TARGET_HEIGHT)
            else:
                resized_frame = frame

            # 2) Optional denoising
            if denoise:
                try:
                    if use_cuda_denoising:
                        gpu_frame = cv2.cuda_GpuMat()
                        gpu_frame.upload(resized_frame)
                        denoised_gpu = cv2.cuda.createFastNlMeansDenoisingColored(
                            gpu_frame, None, 3, 3, 7
                        )
                        denoised_frame = denoised_gpu.download()
                    else:
                        denoised_frame = cv2.fastNlMeansDenoisingColored(
                            resized_frame, None, 3, 3, 7, 21
                        )
                except cv2.error as e:
                    main_logger.error(f"Error applying denoising: {e}")
                    denoised_frame = resized_frame
            else:
                denoised_frame = resized_frame

            # 3) Run YOLO inference (only detecting person=class0 here)
            results = model.predict(
                source=denoised_frame,
                conf=conf_threshold,
                classes=[0],
                verbose=False
            )
            detections = results[0].boxes.data.cpu().numpy()

            # 4) If we have any raw YOLO detections:
            if detections.size > 0:
                # If previously not detecting, set flags
                if not detecting_human:
                    detecting_human = True
                    detection_ongoing.set()
                    detection_count += 1
                    main_logger.info(f"Detections found: {detections}")

                # We track if at least one detection is “kept”
                any_kept_detection = False
                full_frame_filename = None

                # 4A) We *do not* immediately queue a full-frame
                #     We'll only do so if at least one detection is kept.
                #     (So we don't waste a full-frame if they're all masked.)
                
                # 5) Per-detection
                timestamp = datetime.now().strftime('%Y-%m-%d %H:%M:%S')
                SAVE_DIR = get_current_save_dir()  # update date-based subdir

                kept_detections_list = []  # For logging, if you like

                for det_idx, det in enumerate(detections):
                    x1, y1, x2, y2, confidence, class_idx = det

                    # 5A) Check masked regions logic:
                    #     If in a masked region with threshold > this detection's confidence => skip
                    if masked_regions:
                        # Unpack both values from apply_masked_regions
                        keep_detection, skip_info = apply_masked_regions(
                            int(x1), int(y1), int(x2), int(y2),
                            float(confidence),
                            masked_regions
                        )
                        if not keep_detection:
                            # Now skip_info is available
                            zone_name = skip_info.get("zone_name", "UnknownZone")
                            required_conf = skip_info.get("required_conf", 1.0)
                            main_logger.info(
                                f"Skipping detection in masked region '{zone_name}': "
                                f"needed≥{required_conf:.2f}, got={confidence:.2f}, "
                                f"coords=({x1:.0f},{y1:.0f})-({x2:.0f},{y2:.0f})"
                            )
                            # do NOT save or alert
                            continue

                    # If we’re here => detection is *kept*
                    any_kept_detection = True

                    # 5B) If we haven't queued the full-frame yet, do so once
                    if SAVE_FULL_FRAMES and (full_frame_filename is None):
                        full_frame_filename = generate_full_frame_filename(detection_count)
                        try:
                            image_save_queue.put(
                                (denoised_frame.copy(), full_frame_filename, 'full_frame'),
                                block=False
                            )
                            main_logger.info("Queued full-frame image for saving.")
                        except Full:
                            main_logger.warning("Image save queue is full. Dropping full-frame image.")
                            full_frame_filename = None

                    # 5C) Prepare detection-area image if desired
                    detection_area_filename = None
                    if SAVE_DETECTION_AREAS:
                        main_logger.info("Queuing detection area image for saving.")
                        margin = DETECTION_AREA_MARGIN
                        x1_m = max(0, int(x1) - margin)
                        y1_m = max(0, int(y1) - margin)
                        x2_m = min(denoised_frame.shape[1], int(x2) + margin)
                        y2_m = min(denoised_frame.shape[0], int(y2) + margin)

                        detection_area = denoised_frame[y1_m:y2_m, x1_m:x2_m]
                        detection_area_filename = generate_detection_area_filename(detection_count, det_idx)
                        try:
                            image_save_queue.put(
                                (detection_area.copy(), detection_area_filename, 'detection_area'),
                                block=False
                            )
                        except Full:
                            main_logger.warning("Image save queue is full. Dropping detection area image.")
                            detection_area_filename = None

                    # 5D) Build detection_info
                    image_filenames = {}
                    if full_frame_filename:
                        image_filenames['full_frame'] = os.path.relpath(
                            os.path.join(SAVE_DIR, full_frame_filename),
                            SAVE_DIR_BASE
                        )
                    if detection_area_filename:
                        image_filenames['detection_area'] = os.path.relpath(
                            os.path.join(SAVE_DIR, detection_area_filename),
                            SAVE_DIR_BASE
                        )

                    detection_info = {
                        'detection_count': detection_count,
                        'frame_count': int(total_frames),
                        'timestamp': timestamp,
                        'coordinates': (int(x1), int(y1), int(x2), int(y2)),
                        'confidence': float(confidence),
                        'image_filenames': image_filenames
                    }

                    with detections_lock:
                        detections_list.appendleft(detection_info)

                    # 5E) Enqueue Telegram alert
                    alert_message = {
                        "detection_count": detection_info['detection_count'],
                        "frame_count": detection_info['frame_count'],
                        "timestamp": detection_info['timestamp'],
                        "coordinates": detection_info['coordinates'],
                        "confidence": detection_info['confidence']
                    }
                    telegram_alerts.queue_alert(alert_message)

                    # 5F) If draw_rectangles => draw
                    if draw_rectangles:
                        ix1 = max(0, int(x1))
                        iy1 = max(0, int(y1))
                        ix2 = min(denoised_frame.shape[1], int(x2))
                        iy2 = min(denoised_frame.shape[0], int(y2))
                        if not headless:
                            cv2.rectangle(
                                denoised_frame,
                                (ix1, iy1), (ix2, iy2),
                                (0, 255, 0), 2
                            )
                            label = f'Person: {confidence:.2f}'
                            cv2.putText(
                                denoised_frame, label,
                                (ix1, max(iy1 - 10, 0)),
                                cv2.FONT_HERSHEY_SIMPLEX, 0.5,
                                (0, 255, 0), 2
                            )
                        main_logger.info(
                            f"Rectangle drawn: {ix1},{iy1},{ix2},{iy2}"
                            f" with conf {confidence:.2f}"
                        )

                # 6) If we had at least one valid detection, log details
                if any_kept_detection:
                    log_detection_details(detections, total_frames, timestamp)
                    # Start TTS in a separate thread if not running
                    if tts_thread is None or not tts_thread.is_alive():
                        tts_stop_event.clear()
                        tts_thread = threading.Thread(target=announce_detection)
                        tts_thread.start()
                else:
                    # Means all detections were masked/skipped
                    main_logger.info(
                        "All detections in this batch were masked or below threshold; "
                        "no saving/alerts triggered."
                    )

            else:
                # 7) No detections => reset flags if previously detecting
                if detecting_human:
                    detecting_human = False
                    detection_ongoing.clear()
                    tts_stop_event.set()  # Stop TTS if no humans now

            # 8) Send to web server or local display
            if ENABLE_WEBSERVER:
                set_output_frame(denoised_frame)
            elif not headless:
                cv2.imshow('Real-time Human Detection', denoised_frame)
                if cv2.waitKey(1) & 0xFF == ord('q'):
                    stop_event.set()
                    break

            # 9) FPS logging
            total_frames += 1
            if time.time() - last_log_time >= 1:
                fps = total_frames / (time.time() - last_log_time)
                main_logger.info(f'Processed {total_frames} frames at {fps:.2f} FPS')
                last_log_time = time.time()
                total_frames = 0

            # 10) Throttle if needed
            if use_process_fps:
                frame_time = time.time() - start_time
                frame_interval = 1.0 / process_fps
                time_to_wait = frame_interval - frame_time
                if time_to_wait > 0:
                    time.sleep(time_to_wait)

    except Exception as e:
        main_logger.error(f"Error in frame_processing_thread: {e}", exc_info=True)
    finally:
        tts_stop_event.set()
        if not headless:
            cv2.destroyAllWindows()

# # // old logic; changed and obsoleted on Jan 4th, 2025 w/ v0.1615
# # // to be cleaned up ...
#
# def frame_processing_thread(frame_queue, stop_event, conf_threshold, draw_rectangles, denoise, process_fps, use_process_fps, detection_ongoing, headless):
#     global last_tts_time, tts_thread, tts_stop_event
#     model = load_model(DEFAULT_MODEL_VARIANT)
#     use_cuda_denoising = cuda_denoising_available()
#     detection_count = 0
#     last_log_time = time.time()
#     total_frames = 0
#     detecting_human = False

#     # Initialize an empty list to store image filenames
#     image_filenames = []

#     if not headless:
#         # Create a named window with the ability to resize
#         cv2.namedWindow('Real-time Human Detection', cv2.WINDOW_NORMAL)

#     try:
#         while not stop_event.is_set() or not frame_queue.empty():
#             if not frame_queue.empty():
#                 frame = frame_queue.get()
#                 start_time = time.time()

#                 if RESCALE_INPUT:
#                     resized_frame = resize_frame(frame, TARGET_HEIGHT)
#                 else:
#                     resized_frame = frame

#                 if denoise:
#                     try:
#                         if use_cuda_denoising:
#                             gpu_frame = cv2.cuda_GpuMat()
#                             gpu_frame.upload(resized_frame)
#                             denoised_gpu = cv2.cuda.createFastNlMeansDenoisingColored(gpu_frame, None, 3, 3, 7)
#                             denoised_frame = denoised_gpu.download()
#                         else:
#                             denoised_frame = cv2.fastNlMeansDenoisingColored(resized_frame, None, 3, 3, 7, 21)
#                     except cv2.error as e:
#                         main_logger.error(f"Error applying denoising: {e}")
#                         denoised_frame = resized_frame
#                 else:
#                     denoised_frame = resized_frame

#                 results = model.predict(source=denoised_frame, conf=conf_threshold, classes=[0], verbose=False)
#                 detections = results[0].boxes.data.cpu().numpy()

#                 if detections.size > 0:
#                     if not detecting_human:
#                         detecting_human = True
#                         detection_ongoing.set()
#                         detection_count += 1
#                         main_logger.info(f"Detections found: {detections}")

#                     image_filenames = []

#                     # Initialize full_frame_filename outside the loop
#                     full_frame_filename = None

#                     # reset the logic for any kept detections
#                     any_kept_detection = False

#                     # Queue full-frame image for saving
#                     if SAVE_FULL_FRAMES:
#                         main_logger.info("Queuing full-frame image for saving.")
#                         full_frame_filename = generate_full_frame_filename(detection_count)
#                         try:
#                             image_save_queue.put((denoised_frame.copy(), full_frame_filename, 'full_frame'), block=False)
#                         except Full:
#                             main_logger.warning("Image save queue is full. Dropping full-frame image.")
#                             full_frame_filename = None

#                     # Assign timestamp here
#                     timestamp = datetime.now().strftime('%Y-%m-%d %H:%M:%S')

#                     # Update SAVE_DIR to the current date-based directory
#                     SAVE_DIR = get_current_save_dir()

#                     for detection_index, detection in enumerate(detections):
#                         x1, y1, x2, y2, confidence, class_idx = detection

#                         # Call apply_masked_regions if we have masked regions loaded
#                         if len(masked_regions) > 0:
#                             keep_detection = apply_masked_regions(
#                                 int(x1), int(y1), int(x2), int(y2),
#                                 float(confidence),
#                                 masked_regions
#                             )
#                             if not keep_detection:
#                                 main_logger.info(
#                                     f"Skipping detection in masked region: "
#                                     f"({x1:.0f},{y1:.0f})-({x2:.0f},{y2:.0f}), conf={confidence:.2f}"
#                                 )
#                                 continue  # skip this detection entirely

#                         # Initialize the image_info dictionary for this detection
#                         image_info = {}

#                         # Include full-frame filename if available
#                         if full_frame_filename:
#                             image_info['full_frame'] = os.path.relpath(os.path.join(SAVE_DIR, full_frame_filename), SAVE_DIR_BASE)

#                         # Save detection area image and get filename
#                         detection_area_filename = None
#                         if SAVE_DETECTION_AREAS:
#                             main_logger.info("Queuing detection area image for saving.")
#                             main_logger.info(f"Applying margin of {DETECTION_AREA_MARGIN} pixels to detection area.")

#                             # Apply margin
#                             x1_margined = max(0, int(x1) - DETECTION_AREA_MARGIN)
#                             y1_margined = max(0, int(y1) - DETECTION_AREA_MARGIN)
#                             x2_margined = min(denoised_frame.shape[1], int(x2) + DETECTION_AREA_MARGIN)
#                             y2_margined = min(denoised_frame.shape[0], int(y2) + DETECTION_AREA_MARGIN)

#                             detection_area = denoised_frame[y1_margined:y2_margined, x1_margined:x2_margined]
#                             detection_area_filename = generate_detection_area_filename(detection_count, detection_index)
#                             try:
#                                 image_save_queue.put((detection_area.copy(), detection_area_filename, 'detection_area'), block=False)
#                             except Full:
#                                 main_logger.warning("Image save queue is full. Dropping detection area image.")
#                                 detection_area_filename = None

#                             if detection_area_filename:
#                                 image_info['detection_area'] = os.path.relpath(os.path.join(SAVE_DIR, detection_area_filename), SAVE_DIR_BASE)

#                         # Build detection_info for this detection
#                         detection_info = {
#                             'detection_count': detection_count,
#                             'frame_count': int(total_frames),
#                             'timestamp': timestamp,
#                             'coordinates': (
#                                 int(x1), int(y1), int(x2), int(y2)
#                             ),
#                             'confidence': float(confidence),
#                             'image_filenames': image_info
#                         }

#                         # Add detection_info to detections_list
#                         with detections_lock:
#                             detections_list.appendleft(detection_info)

#                         # Construct the detection info as a dictionary
#                         alert_message = {
#                             "detection_count": detection_info.get('detection_count', 0),
#                             "frame_count": detection_info.get('frame_count', 0),
#                             "timestamp": detection_info.get('timestamp', ""),
#                             "coordinates": detection_info.get('coordinates', (0, 0, 0, 0)),
#                             "confidence": detection_info.get('confidence', 0.0)
#                         }

#                         # Queue the alert message as a dictionary
#                         telegram_alerts.queue_alert(alert_message)

#                         if draw_rectangles:
#                             x1_rect, y1_rect = max(0, int(x1)), max(0, int(y1))
#                             x2_rect = min(denoised_frame.shape[1], int(x2))
#                             y2_rect = min(denoised_frame.shape[0], int(y2))
#                             if not headless:
#                                 cv2.rectangle(denoised_frame, (x1_rect, y1_rect), (x2_rect, y2_rect), (0, 255, 0), 2)
#                                 label = f'Person: {confidence:.2f}'
#                                 cv2.putText(denoised_frame, label, (x1_rect, y1_rect - 10), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 255, 0), 2)
#                             main_logger.info(f"Rectangle drawn: {x1_rect}, {y1_rect}, {x2_rect}, {y2_rect} with confidence {confidence:.2f}")

#                     # Log the detection details
#                     log_detection_details(detections, total_frames, timestamp)

#                     # Start the TTS announcement in a separate thread if not already running
#                     if tts_thread is None or not tts_thread.is_alive():
#                         tts_stop_event.clear()
#                         tts_thread = threading.Thread(target=announce_detection)
#                         tts_thread.start()

#                 else:
#                     if detecting_human:
#                         detecting_human = False
#                         detection_ongoing.clear()
#                         tts_stop_event.set()  # Stop TTS when no human is detected

#                 if ENABLE_WEBSERVER:
#                     # Send the processed frame to the web server
#                     set_output_frame(denoised_frame)
#                 elif not headless:
#                     # Display the denoised frame
#                     cv2.imshow('Real-time Human Detection', denoised_frame)
#                     if cv2.waitKey(1) & 0xFF == ord('q'):
#                         stop_event.set()
#                         break

#                 total_frames += 1
#                 if time.time() - last_log_time >= 1:
#                     fps = total_frames / (time.time() - last_log_time)
#                     main_logger.info(f'Processed {total_frames} frames at {fps:.2f} FPS')
#                     last_log_time = time.time()
#                     total_frames = 0

#                 if use_process_fps:
#                     frame_time = time.time() - start_time
#                     frame_interval = 1.0 / process_fps
#                     time_to_wait = frame_interval - frame_time
#                     if time_to_wait > 0:
#                         time.sleep(time_to_wait)
#             else:
#                 time.sleep(0.01)  # Sleep briefly when queue is empty to prevent high CPU usage
#     except Exception as e:
#         main_logger.error(f"Error in frame_processing_thread: {e}", exc_info=True)
#     finally:
#         tts_stop_event.set()  # Ensure TTS stops if the program is stopping
#         if not headless:
#             cv2.destroyAllWindows()

# When the user wants to exit
HEADLESS = False  # Initialize globally

# Define the image-saving thread function
def image_saving_thread_function(image_save_queue, stop_event):
    main_logger.info("Image saving thread started")
    while not stop_event.is_set() or not image_save_queue.empty():
        # Log queue size
        queue_size = image_save_queue.qsize()
        if queue_size > IMAGE_SAVE_QUEUE_MAXSIZE * 0.8:
            main_logger.warning(f"Image save queue size is at {queue_size}/{IMAGE_SAVE_QUEUE_MAXSIZE}")

        try:
            frame, filename, image_type = image_save_queue.get(timeout=1)
            main_logger.info(f"Received image to save: {image_type}, filename: {filename}")

            filepath = None
            if image_type == 'full_frame':
                try:
                    filepath = save_full_frame_image(frame, filename)
                except Exception as e:
                    main_logger.error(f"Error during full-frame image saving: {e}")
            elif image_type == 'detection_area':
                try:
                    filepath = save_detection_area_image(frame, filename)
                except Exception as e:
                    main_logger.error(f"Error during detection area image saving: {e}")
            else:
                main_logger.warning(f"Unknown image type: {image_type}")

            if filepath:
                # Enqueue file for remote sync if enabled
                if remote_sync_obj.REMOTE_SYNC_ENABLED:
                    try:
                        remote_sync_obj.enqueue_file(filepath)
                    except Full:
                        main_logger.warning(f"Remote sync queue is full. Dropping file: {filepath}")
        except Empty:
            continue
        except Exception as e:
            main_logger.error(f"Error in image_saving_thread: {e}")
    main_logger.info("Image saving thread stopped")

# Start the image-saving thread
image_saving_thread = threading.Thread(
    target=image_saving_thread_function,
    args=(image_save_queue, image_saving_stop_event)
)
image_saving_thread.start()

# save full frame image
def save_full_frame_image(frame, filename):
    global SAVE_DIR, SAVE_DIR_BASE
    try:
        SAVE_DIR = get_current_save_dir()
        filepath = os.path.join(SAVE_DIR, filename)
        main_logger.info(f"Attempting to save full-frame image to: {filepath}")

        # Set image saving parameters
        params = []
        if FULL_FRAME_IMAGE_FORMAT.lower() in ['jpg', 'jpeg']:
            params = [cv2.IMWRITE_JPEG_QUALITY, IMAGE_QUALITY]
        elif FULL_FRAME_IMAGE_FORMAT.lower() == 'webp':
            if WEBP_LOSSLESS:
                params = [cv2.IMWRITE_WEBP_QUALITY, 101]
            else:
                params = [cv2.IMWRITE_WEBP_QUALITY, IMAGE_QUALITY]
        elif FULL_FRAME_IMAGE_FORMAT.lower() == 'png':
            params = [cv2.IMWRITE_PNG_COMPRESSION, PNG_COMPRESSION_LEVEL]

        if not cv2.imwrite(filepath, frame, params):
            main_logger.error(f"Failed to save full-frame image: {filepath}")
            return None
        else:
            main_logger.info(f"Saved full-frame image: {filepath}")
            return filepath
    except Exception as e:
        main_logger.error(f"Error saving full-frame image: {e}")
        return None
    
# save detection area image
def save_detection_area_image(frame, filename):
    global SAVE_DIR, SAVE_DIR_BASE
    try:
        SAVE_DIR = get_current_save_dir()
        filepath = os.path.join(SAVE_DIR, filename)
        main_logger.info(f"Attempting to save detection area image to: {filepath}")

        # Set image saving parameters
        params = []
        if DETECTION_AREA_IMAGE_FORMAT.lower() in ['jpg', 'jpeg']:
            params = [cv2.IMWRITE_JPEG_QUALITY, IMAGE_QUALITY]
        elif DETECTION_AREA_IMAGE_FORMAT.lower() == 'webp':
            if WEBP_LOSSLESS:
                params = [cv2.IMWRITE_WEBP_QUALITY, 101]
            else:
                params = [cv2.IMWRITE_WEBP_QUALITY, IMAGE_QUALITY]
        elif DETECTION_AREA_IMAGE_FORMAT.lower() == 'png':
            params = [cv2.IMWRITE_PNG_COMPRESSION, PNG_COMPRESSION_LEVEL]

        if not cv2.imwrite(filepath, frame, params):
            main_logger.error(f"Failed to save detection area image: {filepath}")
            return None
        else:
            main_logger.info(f"Saved detection area image: {filepath}")
            return filepath
    except Exception as e:
        main_logger.error(f"Error saving detection area image: {e}")
        return None

# helper functions for image saving
def generate_full_frame_filename(detection_count):
    timestamp = datetime.now().strftime("%Y%m%d-%H%M%S")
    filename = f"{timestamp}_{detection_count}.{FULL_FRAME_IMAGE_FORMAT}"
    return filename

def generate_detection_area_filename(detection_count, detection_index):
    timestamp = datetime.now().strftime("%Y%m%d-%H%M%S")
    filename = f"{timestamp}_{detection_count}_area.{DETECTION_AREA_IMAGE_FORMAT}"
    # filename = f"{timestamp}_{detection_count}_det{detection_index}.{DETECTION_AREA_IMAGE_FORMAT}"
    return filename

def signal_handler(sig, frame):
    main_logger.info("Interrupt received, stopping, please wait for the program to finish...")
    stop_event.set()  # Signal threads to stop
    image_saving_stop_event.set()  # Signal the image saving thread to stop
    image_saving_thread.join() 

    # Stop remote sync thread
    remote_sync_obj.stop()

    if not HEADLESS and not ENABLE_WEBSERVER:
        cv2.destroyAllWindows()
    sys.exit(0)

# Making sure the data is compliant
def sanitize_detection_data(detection):
    return {
        'frame_count': int(detection['frame_count']),
        'timestamp': str(detection['timestamp']),
        'coordinates': [int(coord) for coord in detection['coordinates']],
        'confidence': float(detection['confidence'])
    }

# Main
if __name__ == "__main__":
    log_cuda_info()  # Log CUDA information
    parser = argparse.ArgumentParser(description="YOLOv8 RTMP Stream Human Detection")

    parser.add_argument("--use_env_save_dir", action='store_true', help="Use environment variable for save directory")
    parser.add_argument("--save_full_frames", action='store_true', default=None, help="Save full-frame images when detections occur")
    parser.add_argument("--no_save_full_frames", action='store_false', dest='save_full_frames', default=None, help="Do not save full-frame images")
    parser.add_argument("--save_detection_areas", action='store_true', default=None, help="Save detection area images")
    parser.add_argument("--no_save_detection_areas", action='store_false', dest='save_detection_areas', default=None, help="Do not save detection area images")
    parser.add_argument("--full_frame_image_format", type=str, help="Image format for full-frame images (jpg or png)")
    parser.add_argument("--detection_area_image_format", type=str, help="Image format for detection area images (jpg or png)")
    parser.add_argument("--detection_area_margin", type=int, help="Margin in pixels to add around the detection area when saving images")
    parser.add_argument("--IMAGE_SAVE_QUEUE_MAXSIZE", type=int, help="Size of the frame queue. Set to 0 for unlimited size.")
    parser.add_argument("--image_quality", type=int, help="Image quality for jpg and webp formats (0-100)")
    parser.add_argument("--webp_lossless", action='store_true', help="Enable lossless compression for WebP images")
    parser.add_argument("--png_compression_level", type=int, help="PNG compression level (0-9)")

    # Define other arguments without default values to allow config.ini to set them
    parser.add_argument("--headless", action='store_true', help="Run in headless mode without GUI display")
    parser.add_argument("--stream_url", type=str, help="URL of the RTMP stream")
    parser.add_argument("--use_webcam", action='store_true', help="Use webcam for video input")
    parser.add_argument("--webcam_index", type=int, help="Index number of the webcam to use")
    parser.add_argument("--conf_threshold", type=float, help="Confidence threshold for detections")
    parser.add_argument("--model_variant", type=str, help="YOLOv8 model variant to use")
    parser.add_argument("--image_format", type=str, help="Image format for saving detections (jpg or png)")
    parser.add_argument("--save_dir", type=str, help="Directory to save detection images")
    parser.add_argument("--retry_delay", type=int, help="Delay between retries to connect to the stream")
    parser.add_argument("--max_retries", type=int, help="Maximum number of retries to connect to the stream")
    parser.add_argument("--rescale_input", action='store_true', help="Rescale the input frames")
    parser.add_argument("--target_height", type=int, help="Target height for rescaling the frames")
    parser.add_argument("--denoise", action='store_true', help="Toggle denoising on/off")
    parser.add_argument("--process_fps", type=int, help="Frames per second for processing")
    parser.add_argument("--use_process_fps", action='store_true', help="Whether to use custom processing FPS")
    parser.add_argument("--timeout", type=int, help="Timeout for the video stream in seconds")
    parser.add_argument("--enable_webserver", action='store_true', help="Enable the web server for streaming detections")
    parser.add_argument("--webserver_host", type=str, help="Host IP address for the web server")
    parser.add_argument("--webserver_port", type=int, help="Port number for the web server")

    args = parser.parse_args()

    # Update configurations based on command-line arguments
    if args.use_env_save_dir:
        USE_ENV_SAVE_DIR = True
    else:
        USE_ENV_SAVE_DIR = config.getboolean('detection', 'use_env_save_dir', fallback=False)

    # Set saved image quality
    if args.image_quality is not None:
        IMAGE_QUALITY = args.image_quality
    main_logger.info(f"Image quality is set to: {IMAGE_QUALITY}")

    # After parsing arguments
    if args.save_full_frames is not None:
        SAVE_FULL_FRAMES = args.save_full_frames
    main_logger.info(f"SAVE_FULL_FRAMES is set to: {SAVE_FULL_FRAMES}")        

    if args.save_detection_areas is not None:
        SAVE_DETECTION_AREAS = args.save_detection_areas
    # main_logger.info(f"SAVE_DETECTION_AREAS is set to: {SAVE_FULL_FRAMES}")                
    main_logger.info(f"SAVE_DETECTION_AREAS is set to: {SAVE_DETECTION_AREAS}")
    
    if args.full_frame_image_format:
        FULL_FRAME_IMAGE_FORMAT = args.full_frame_image_format
    if args.detection_area_image_format:
        DETECTION_AREA_IMAGE_FORMAT = args.detection_area_image_format
    if args.detection_area_margin is not None:
        DETECTION_AREA_MARGIN = args.detection_area_margin
    main_logger.info(f"DETECTION_AREA_MARGIN is set to: {DETECTION_AREA_MARGIN}")
    if args.IMAGE_SAVE_QUEUE_MAXSIZE is not None:
        IMAGE_SAVE_QUEUE_MAXSIZE = args.IMAGE_SAVE_QUEUE_MAXSIZE
        if IMAGE_SAVE_QUEUE_MAXSIZE <= 0:
            main_logger.info("Frame queue size set to unlimited via command-line argument.")
            IMAGE_SAVE_QUEUE_MAXSIZE = 0  # Queue with unlimited size
        else:
            main_logger.info(f"Frame queue size set to {IMAGE_SAVE_QUEUE_MAXSIZE} via command-line argument.")

    # Override configurations based on command-line arguments
    if args.enable_webserver:
        ENABLE_WEBSERVER = True
    if args.webserver_host:
        WEBSERVER_HOST = args.webserver_host
    if args.webserver_port:
        WEBSERVER_PORT = args.webserver_port

    # Override HEADLESS if --headless is specified
    if args.headless or ENABLE_WEBSERVER:
        HEADLESS = True
        main_logger.info("Running in headless mode -- no GUI. Set 'headless' AND 'enable_webserver' to 'false' if you want to run the windowed GUI version.")

    # # Set SAVE_DETECTIONS based on args, with fallback to config
    # # After parsing arguments
    # if args.save_detections is not None:
    #     SAVE_DETECTIONS = args.save_detections
    #     main_logger.info(f"save_detections overridden by command-line argument: {SAVE_DETECTIONS}")
    # else:
    #     SAVE_DETECTIONS = config.getboolean('detection', 'save_detections')
    #     main_logger.info(f"save_detections set from config.ini: {SAVE_DETECTIONS}")

    # # Similarly, set DRAW_RECTANGLES based on args, with fallback to config
    # if args.draw_rectangles is not None:
    #     DRAW_RECTANGLES = args.draw_rectangles
    # else:
    DRAW_RECTANGLES = config.getboolean('detection', 'draw_rectangles')
    main_logger.info(f"Draw rectangles set to: {DRAW_RECTANGLES}")

    # Update other configurations based on arguments, only if provided
    if args.stream_url:
        STREAM_URL = args.stream_url
    if args.use_webcam:
        USE_WEBCAM = args.use_webcam
    if args.webcam_index is not None:
        WEBCAM_INDEX = args.webcam_index
    if args.conf_threshold is not None:
        DEFAULT_CONF_THRESHOLD = args.conf_threshold
    if args.model_variant:
        DEFAULT_MODEL_VARIANT = args.model_variant
    if args.image_format:
        IMAGE_FORMAT = args.image_format
    if args.save_dir:
        SAVE_DIR = args.save_dir
    if args.retry_delay is not None:
        RETRY_DELAY = args.retry_delay
    if args.max_retries is not None:
        MAX_RETRIES = args.max_retries
    if args.rescale_input:
        RESCALE_INPUT = args.rescale_input
    if args.target_height is not None:
        TARGET_HEIGHT = args.target_height
    if args.denoise:
        DENOISE = args.denoise
    if args.process_fps is not None:
        PROCESS_FPS = args.process_fps
    if args.use_process_fps:
        USE_PROCESS_FPS = args.use_process_fps
    if args.timeout is not None:
        TIMEOUT = args.timeout
    if args.webp_lossless:
        WEBP_LOSSLESS = True

    # Initialize web server if enabled
    if ENABLE_WEBSERVER:
        # Start the web server in a separate daemon thread
        web_server_thread = threading.Thread(
            target=start_web_server,
            args=(
                WEBSERVER_HOST,        # host
                WEBSERVER_PORT,        # port
                detection_log_path,    # detection_log_path
                detections_list,       # detections_list
                logs_list,             # logs_list
                detections_lock,       # detections_lock
                logs_lock,             # logs_lock
                config,                 # config
                SAVE_DIR_BASE          # Pass SAVE_DIR_BASE explicitly                
                # detection_timestamps,  # Pass detection_timestamps
                # timestamps_lock        # Pass timestamps_lock
            ),
            daemon=True  # Set as daemon
        )
        web_server_thread.start()
        main_logger.info(f"Web server started at http://{WEBSERVER_HOST}:{WEBSERVER_PORT}")

    # Now, load configurations
    config = load_config()

    # Initialize SAVE_DIR_BASE
    SAVE_DIR_BASE = get_base_save_dir()
    main_logger.info(f"SAVE_DIR_BASE is set to: {SAVE_DIR_BASE}")

    # Initialize SAVE_DIR using the updated get_save_dir function
    try:
        SAVE_DIR = get_current_save_dir()
        main_logger.info(f"Saving detections to directory: {SAVE_DIR}")
    except RuntimeError as e:
        main_logger.error(f"Error initializing save directory: {e}")
        sys.exit(1)

    # Checking variables before taking off...
    if DETECTION_AREA_MARGIN < 0:
        main_logger.error("DETECTION_AREA_MARGIN is negative. Setting it to 0.")
        DETECTION_AREA_MARGIN = 0

    # Handle Unlimited Queue Size
    if IMAGE_SAVE_QUEUE_MAXSIZE <= 0:
        main_logger.warning("Frame queue size set to unlimited. This may cause bottlenecks depending on your setup!")
        IMAGE_SAVE_QUEUE_MAXSIZE = 0  # Queue with unlimited size
    else:
        main_logger.info(f"Frame saving queue maximum size set to: {IMAGE_SAVE_QUEUE_MAXSIZE}")

    if FULL_FRAME_IMAGE_FORMAT.lower() not in SUPPORTED_IMAGE_FORMATS:
        main_logger.error(f"Unsupported full-frame image format: {FULL_FRAME_IMAGE_FORMAT}. Defaulting to 'jpg'.")
        FULL_FRAME_IMAGE_FORMAT = 'jpg'

    if DETECTION_AREA_IMAGE_FORMAT.lower() not in SUPPORTED_IMAGE_FORMATS:
        main_logger.error(f"Unsupported detection area image format: {DETECTION_AREA_IMAGE_FORMAT}. Defaulting to 'jpg'.")
        DETECTION_AREA_IMAGE_FORMAT = 'jpg'

    if args.png_compression_level is not None:
        PNG_COMPRESSION_LEVEL = args.png_compression_level
        main_logger.info(f"PNG compression level overridden by command-line argument: {PNG_COMPRESSION_LEVEL}")

    if PNG_COMPRESSION_LEVEL < 0 or PNG_COMPRESSION_LEVEL > 9:
        main_logger.warning("PNG compression level must be between 0 and 9. Setting to default value of 3.")
        PNG_COMPRESSION_LEVEL = 3

    if FULL_FRAME_IMAGE_FORMAT.lower() == 'png':
        main_logger.info(f"Full-frame images will be saved as PNG with compression level {PNG_COMPRESSION_LEVEL}.")
    if DETECTION_AREA_IMAGE_FORMAT.lower() == 'png':
        main_logger.info(f"Detection area images will be saved as PNG with compression level {PNG_COMPRESSION_LEVEL}.")

    if FULL_FRAME_IMAGE_FORMAT.lower() == 'webp' and WEBP_LOSSLESS:
        main_logger.info("Full-frame images will be saved as lossless WebP.")
    elif FULL_FRAME_IMAGE_FORMAT.lower() == 'webp':
        main_logger.info(f"Full-frame images will be saved as lossy WebP with quality {IMAGE_QUALITY}.")

    if DETECTION_AREA_IMAGE_FORMAT.lower() == 'webp' and WEBP_LOSSLESS:
        main_logger.info("Detection area images will be saved as lossless WebP.")
    elif DETECTION_AREA_IMAGE_FORMAT.lower() == 'webp':
        main_logger.info(f"Detection area images will be saved as lossy WebP with quality {IMAGE_QUALITY}.")

    main_logger.info(f"Full-frame image format: {FULL_FRAME_IMAGE_FORMAT}")
    main_logger.info(f"Detection area image format: {DETECTION_AREA_IMAGE_FORMAT}")
    main_logger.info(f"Image quality: {IMAGE_QUALITY}")
    main_logger.info(f"PNG compression level is set to: {PNG_COMPRESSION_LEVEL}")    
    main_logger.info(f"WebP lossless compression is {'enabled' if WEBP_LOSSLESS else 'disabled'}")

    # Initialize queues and events
    frame_queue = Queue(maxsize=IMAGE_SAVE_QUEUE_MAXSIZE)
    stop_event = Event()
    detection_ongoing = Event()

    # Register signal handler for clean exit
    signal.signal(signal.SIGINT, signal_handler)
    signal.signal(signal.SIGTERM, signal_handler)

    # Initialize threads
    capture_thread = Thread(
    target=frame_capture_thread,
    args=(STREAM_URL, USE_WEBCAM, WEBCAM_INDEX, frame_queue, stop_event)
    )
    processing_thread = Thread(target=frame_processing_thread, args=(frame_queue, stop_event, DEFAULT_CONF_THRESHOLD, DRAW_RECTANGLES, DENOISE, PROCESS_FPS, USE_PROCESS_FPS, detection_ongoing, HEADLESS))

    try:
        capture_thread.start()
        processing_thread.start()

        capture_thread.join()
        processing_thread.join()

    finally:
        # Ensure TTS thread is stopped
        tts_stop_event.set()
        if tts_thread is not None:
            tts_thread.join()
        main_logger.info("Cleaning up threads and resources.")
        if not HEADLESS and not ENABLE_WEBSERVER:
            cv2.destroyAllWindows()
        main_logger.info("Program exited cleanly.")