import cv2
import numpy as np
from ultralytics import YOLO
import os
import uuid

# Load YOLOv8 model (can be customized for object detection)
model = YOLO("yolov8n.pt")  # You may switch to yolov8s.pt or yolov8m.pt based on size/accuracy

# Maintain state of detected objects for simple tracking
unattended_objects_memory = {}

def detect_unattended_objects(frame, frame_index, detection_interval=30, stay_threshold=90):
    """
    Detects unattended objects such as bags that appear in the same position for too long.
    Returns True and the cropped object image if detected.
    """

    detections = model.predict(source=frame, classes=[24, 26, 28, 39], verbose=False)[0]
    # COCO classes for suitcase (24), handbag (26), backpack (28), and cardboard box (39)

    h, w, _ = frame.shape
    detected = False
    cropped_output = None

    for result in detections.boxes:
        cls = int(result.cls[0])
        conf = float(result.conf[0])
        x1, y1, x2, y2 = map(int, result.xyxy[0])

        center = ((x1 + x2) // 2, (y1 + y2) // 2)
        object_id = f"{cls}-{center[0]//20}-{center[1]//20}"  # simple spatial bin id

        if object_id not in unattended_objects_memory:
            unattended_objects_memory[object_id] = frame_index
        else:
            duration = frame_index - unattended_objects_memory[object_id]
            if duration >= stay_threshold:
                detected = True
                cropped_output = frame[y1:y2, x1:x2]
                unattended_objects_memory[object_id] = frame_index + 9999  # suppress repeated detection

    # Clear memory periodically
    if frame_index % (stay_threshold * 2) == 0:
        unattended_objects_memory.clear()

    return detected, cropped_output