"""
Murderer Detector - A Humorous Webcam Person Detection Demo

This app demonstrates real-time person detection with humorous labeling.
It's structured to be easily modified for serious applications like:
- Security monitoring
- PPE detection
- Customer analytics
- Social distancing monitoring

Architecture:
1. Detection Module: Uses YOLO for person detection (easily swappable)
2. Classification Logic: Generates humorous labels (swap for real ML inference)
3. Annotation Layer: Draws boxes and labels (customize visuals)
4. Streaming Handler: Processes webcam feed via Gradio
"""

import gradio as gr
import cv2
import numpy as np
from ultralytics import YOLO
import random
from typing import Tuple, List, Dict
import time
import os


# ============================================================================
# DETECTION MODULE
# Swap this section for different detection models or backends
# ============================================================================

class PersonDetector:
    """
    Person detection using YOLO.

    For serious applications, modify this to:
    - Use different models (MediaPipe, custom trained models)
    - Add specific object detection (weapons, PPE, etc.)
    - Integrate with cloud APIs (AWS Rekognition, Google Vision)
    """

    def __init__(self, model_name: str = "yolov8n.pt", confidence: float = 0.5):
        """
        Initialize the person detector.

        Args:
            model_name: YOLO model to use (n=nano, s=small, m=medium, l=large)
            confidence: Detection confidence threshold
        """
        self.model = YOLO(model_name)
        self.confidence = confidence

    def detect_persons(self, frame: np.ndarray) -> List[Dict]:
        """
        Detect persons in a frame.

        Args:
            frame: Input image as numpy array (BGR format)

        Returns:
            List of detections with bounding boxes and confidence scores
        """
        # Run inference
        results = self.model(frame, conf=self.confidence,
                             classes=[0], verbose=False)

        detections = []
        for result in results:
            boxes = result.boxes
            for box in boxes:
                x1, y1, x2, y2 = box.xyxy[0].cpu().numpy()
                confidence = float(box.conf[0])

                detections.append({
                    'bbox': (int(x1), int(y1), int(x2), int(y2)),
                    'confidence': confidence
                })

        return detections


# ============================================================================
# CLASSIFICATION LOGIC
# Replace this section for serious applications
# ============================================================================

class MurdererClassifier:
    """
    Humorous 'threat' classification.

    For serious applications, replace with:
    - Emotion detection models
    - Action recognition (violence, falls, etc.)
    - Anomaly detection
    - Age/gender classification
    """

    # Short, punchy threat labels
    LABELS = [
        "🥣 SERIAL BREAKFAST SKIPPER",
        "😴 SUSPICIOUSLY WELL-RESTED",
        "📺 TRUE CRIME WATCHER",
        "📚 DANGEROUS BOOK READER",
        "🥄 SERIAL CEREAL KILLER",
        "😊 TOO POLITE. SUSPICIOUS.",
        "👀 PROFESSIONAL LURKER",
        "🪴 PLANT WHISPERER",
        "🐕 ALLEGED DOG PETTER",
        "☕ NOTORIOUS TEA DRINKER",
        "🤔 CONFIRMED OVERTHINKER",
        "🧍 BACKGROUND STANDER",
        "💤 GETS 8 HOURS SLEEP",
        "💧 DRINKS WATER DAILY",
        "🔌 OWNS MULTIPLE USB-C CABLES",
        "🎵 CAN WHISTLE & SNAP",
        "🛏️ FITTED SHEET FOLDER",
        "📖 READ 3+ BOOKS",
        "🌿 TALKS TO HOUSEPLANTS",
        "🎯 UNUSUALLY GOOD AT TRIVIA",
    ]

    def __init__(self):
        """Initialize the classifier with tracking for consistent labels."""
        self.person_history = {}
        self.next_id = 0

    def classify(self, detection: Dict) -> Dict:
        """
        Generate humorous classification for a detected person.

        Args:
            detection: Detection dict with bbox and confidence

        Returns:
            Classification dict with label and threat level
        """
        # Generate random threat assessment
        threat_level = random.randint(45, 99)
        label = random.choice(self.LABELS)

        return {
            'threat_level': threat_level,
            'label': label,
            'confidence': detection['confidence']
        }


# ============================================================================
# ANNOTATION LAYER
# Customize this section for different visual styles
# ============================================================================

class FrameAnnotator:
    """
    Draws annotations on frames.

    Modify this to:
    - Change colors, styles, fonts
    - Add different visualization modes
    - Include overlay graphics or warnings
    - Show statistics or heatmaps
    """

    def __init__(self):
        """Initialize annotator with color schemes."""
        self.colors = {
            'high_threat': (0, 0, 255),    # Red
            'medium_threat': (0, 165, 255),  # Orange
            'low_threat': (0, 255, 255),    # Yellow
            'text_bg': (0, 0, 0),           # Black
            'text': (255, 255, 255)         # White
        }

    def get_threat_color(self, threat_level: int) -> Tuple[int, int, int]:
        """Get color based on threat level."""
        if threat_level >= 80:
            return self.colors['high_threat']
        elif threat_level >= 60:
            return self.colors['medium_threat']
        else:
            return self.colors['low_threat']

    def annotate_frame(
        self,
        frame: np.ndarray,
        detections: List[Dict],
        classifications: List[Dict]
    ) -> np.ndarray:
        """
        Draw annotations on frame.

        Args:
            frame: Input frame
            detections: List of detection dicts
            classifications: List of classification dicts

        Returns:
            Annotated frame
        """
        annotated = frame.copy()

        # Draw header
        self._draw_header(annotated, len(detections))

        # Annotate each detection
        for detection, classification in zip(detections, classifications):
            self._draw_detection(annotated, detection, classification)

        return annotated

    def _draw_header(self, frame: np.ndarray, num_suspects: int):
        """Draw header with suspect count."""
        header_text = f"🚨 SUSPECTS DETECTED: {num_suspects} 🚨"
        font = cv2.FONT_HERSHEY_DUPLEX
        font_scale = 0.8
        thickness = 2

        # Get text size
        (text_width, text_height), baseline = cv2.getTextSize(
            header_text, font, font_scale, thickness
        )

        # Draw background
        cv2.rectangle(
            frame,
            (0, 0),
            (frame.shape[1], text_height + baseline + 20),
            (0, 0, 0),
            -1
        )

        # Draw text
        x = (frame.shape[1] - text_width) // 2
        y = text_height + 10
        cv2.putText(
            frame, header_text, (x, y),
            font, font_scale, (0, 0, 255), thickness
        )

    def _draw_detection(
        self,
        frame: np.ndarray,
        detection: Dict,
        classification: Dict
    ):
        """Draw bounding box and labels for a detection."""
        x1, y1, x2, y2 = detection['bbox']
        threat_level = classification['threat_level']
        color = self.get_threat_color(threat_level)

        # Draw bounding box
        cv2.rectangle(frame, (x1, y1), (x2, y2), color, 3)

        # Create single label with threat level
        label = f"{classification['label']} ({threat_level}%)"

        # Draw label with larger, more readable text
        font = cv2.FONT_HERSHEY_DUPLEX
        font_scale = 0.7
        thickness = 2
        padding = 8

        (text_width, text_height), baseline = cv2.getTextSize(
            label, font, font_scale, thickness
        )

        # Position label above bounding box, or below if too close to top
        y_offset = y1 - 10
        if y_offset - text_height - padding < 0:
            y_offset = y2 + text_height + padding + 10

        # Draw background rectangle
        cv2.rectangle(
            frame,
            (x1, y_offset - text_height - padding),
            (x1 + text_width + padding * 2, y_offset + baseline + padding),
            self.colors['text_bg'],
            -1
        )

        # Draw text
        cv2.putText(
            frame,
            label,
            (x1 + padding, y_offset),
            font,
            font_scale,
            color,
            thickness
        )


# ============================================================================
# STREAMING HANDLER
# Main processing pipeline
# ============================================================================

class MurdererDetector:
    """
    Main application class that combines all modules.
    """

    def __init__(self):
        """Initialize all components."""
        print("🔍 Initializing Murderer Detector...")
        self.detector = PersonDetector(model_name="yolov8n.pt", confidence=0.5)
        self.classifier = MurdererClassifier()
        self.annotator = FrameAnnotator()
        self.previous_suspect_count = 0
        print("✅ Ready to detect suspicious individuals!")

    def process_frame(self, frame: np.ndarray) -> Tuple[np.ndarray, int]:
        """
        Process a single frame.

        Args:
            frame: Input frame from webcam

        Returns:
            Tuple of (annotated frame, suspect count)
        """
        if frame is None:
            return None, 0

        # Detect all persons
        all_detections = self.detector.detect_persons(frame)

        # Filter out the user (largest person, presumably closest to camera)
        # Only flag people behind the user
        detections = self._filter_user(all_detections)

        # Classify each detection
        classifications = [
            self.classifier.classify(det) for det in detections
        ]

        # Annotate frame
        annotated = self.annotator.annotate_frame(
            frame, detections, classifications
        )

        # Track count for sound alerts
        current_count = len(detections)

        return annotated, current_count

    def _filter_user(self, detections: List[Dict]) -> List[Dict]:
        """
        Filter out the user (largest person) from detections.

        The assumption is that the user is sitting in front of the webcam
        and will be the largest person in the frame. We want to flag only
        people behind them.

        Args:
            detections: List of all person detections

        Returns:
            Filtered detections excluding the user
        """
        if len(detections) <= 1:
            # If only one person or no people, don't flag anyone
            return []

        # Calculate area for each detection
        detections_with_area = []
        for det in detections:
            x1, y1, x2, y2 = det['bbox']
            area = (x2 - x1) * (y2 - y1)
            detections_with_area.append((det, area))

        # Sort by area (largest first)
        detections_with_area.sort(key=lambda x: x[1], reverse=True)

        # Return all except the largest (the user)
        return [det for det, _ in detections_with_area[1:]]


# ============================================================================
# GRADIO INTERFACE
# ============================================================================

def create_interface():
    """Create and configure the Gradio interface."""

    # Initialize detector
    app = MurdererDetector()

    # Create interface
    with gr.Blocks(title="Murderer Detector 🔪") as demo:

        gr.Markdown("""
        # 🔪 Murderer Detector 🔍

        **DISCLAIMER:** This is for funs.
                    """)

        # Unified webcam display (input and output combined)
        webcam = gr.Image(
            sources="webcam",
            streaming=True,
            label="Murderer Detector (requires webcam on)"
        )

        # Hidden counter for sound alerts (triggers JS when value changes)
        suspect_count = gr.Number(value=0, visible=False)

        # Sound alert JavaScript
        alert_sound = gr.HTML("""
        <script>
        let previousCount = 0;
        const audioContext = new (window.AudioContext || window.webkitAudioContext)();

        function playAlertSound() {
            // Create funny "ding-dong" doorbell sound
            const now = audioContext.currentTime;
            const oscillator1 = audioContext.createOscillator();
            const oscillator2 = audioContext.createOscillator();
            const gainNode = audioContext.createGain();

            // First "ding" (higher note)
            oscillator1.connect(gainNode);
            oscillator1.frequency.value = 800;
            oscillator1.type = 'sine';

            // Second "dong" (lower note)
            oscillator2.connect(gainNode);
            oscillator2.frequency.value = 600;
            oscillator2.type = 'sine';

            gainNode.connect(audioContext.destination);
            gainNode.gain.setValueAtTime(0.3, now);
            gainNode.gain.exponentialRampToValueAtTime(0.01, now + 0.15);

            // Play ding
            oscillator1.start(now);
            oscillator1.stop(now + 0.15);

            // Play dong
            oscillator2.start(now + 0.15);
            oscillator2.stop(now + 0.3);
        }

        // Monitor for count changes
        setInterval(() => {
            const countElement = document.querySelector('input[type="number"]');
            if (countElement) {
                const currentCount = parseInt(countElement.value) || 0;
                if (currentCount > previousCount && currentCount > 0) {
                    playAlertSound();
                    console.log('🚨 New suspect detected! Count:', currentCount);
                }
                previousCount = currentCount;
            }
        }, 100);
        </script>
        """)

        gr.Markdown("""
        # 📊 How It Works

        1. ** Enable your webcam ** - Click the webcam button above, then Record
        2. ** Detects creepers behind you **
        3. ** Plays alert sound ** when new suspects appear! 🔊

        # 🎯 Detection Features

        - Real-time person detection using YOLOv8
        - Threat level assessment(totally scientific 😅)
        - Color-coded danger ratings
        - Running suspect count
        - Sound alerts (ding-dong when count increases!)
        """)

        # Set up streaming (unified input/output)
        webcam.stream(
            fn=app.process_frame,
            inputs=[webcam],
            outputs=[webcam, suspect_count],
            stream_every=0.1,
            concurrency_limit=30
        )

    return demo


# ============================================================================
# MAIN
# ============================================================================

if __name__ == "__main__":
    demo = create_interface()
    # Only use share=True for local development (not on HF Spaces)
    is_spaces = os.getenv("SPACE_ID") is not None
    demo.launch(share=not is_spaces)