🌐 Live Website: https://liftproof-ai-website-7oediq4sy-md-s-projects-90825e24.vercel.app/

LiftProof.Ai is an advanced real-time shoplifting detection system that uses YOLOv8-Pose for skeleton tracking combined with behavioral analysis algorithms to identify potential theft in retail environments. Unlike traditional object detection systems, LiftProof.Ai analyzes human behavior patterns to detect concealment actions before items leave the store.

🎯 Problem It Solves

• $100+ billion lost annually to retail theft globally
• Traditional CCTV requires constant human monitoring
• Object-based detection fails to catch concealment behaviors
• Existing solutions are expensive and complex

💡 Our Solution

LiftProof.Ai provides:

• Real-time behavioral analysis using pose estimation
• Multi-camera support for major CCTV brands
• Instant alerts to security personnel
• Evidence capture with timestamps and video clips
• Cost-effective deployment on standard hardware

---

✨ Features

Core Detection Capabilities

Feature	Description
🖐️ Pocket Concealment	Detects hand-to-pocket movements indicating item hiding
👕 Under-Shirt Hiding	Identifies items being tucked under clothing
🩳 Waistband Concealment	Monitors waistband area for hiding behavior
⚡ Quick Grab Detection	Catches rapid snatching movements
👀 Nervous Behavior	Tracks suspicious head movements and looking around

Technical Features

• ✅ Real-time Processing - 30+ FPS on modern hardware
• ✅ Multi-Person Tracking - Track multiple subjects simultaneously
• ✅ Skeleton Visualization - Full 17-keypoint pose estimation
• ✅ Progressive Alerts - Normal → Watching → Suspicious → Alert
• ✅ Suspicion Scoring - 0-100% risk assessment per person
• ✅ Evidence Capture - Automatic screenshot/video on alert
• ✅ CCTV Integration - Works with Hikvision, Dahua, Lorex, UNV, and more

---

🚀 Installation

Prerequisites

• Python 3.8 or higher
• Webcam or CCTV camera (RTSP compatible)
• macOS, Windows, or Linux

Quick Install

# Clone the repository
git clone https://github.com/Mrabbi3/LiftProofAI.git
cd LiftProofAI

# Create virtual environment
python3 -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

Requirements

ultralytics>=8.1.0
opencv-python>=4.8.0
numpy>=1.24.0
torch>=2.0.0

Download YOLOv8-Pose Model

The model downloads automatically on first run, or manually:

python3 -c "from ultralytics import YOLO; YOLO('yolov8n-pose.pt')"

---

📖 Usage

Basic Usage (Webcam)

# Activate virtual environment
source venv/bin/activate

# Run detection
python3 liftproof_v2.py

With CCTV Camera

python3 main.py --camera rtsp://admin:password@192.168.1.64:554/stream1

Testing the System

Once running, test these behaviors:

Action	Expected Response
Stand normally	🟢 Green skeleton - "NORMAL"
Hand in pocket (hold 1-2 sec)	🟡→🟠→🔴 Alert triggered
Hand to waistband	🟡→🟠→🔴 Alert triggered
Hand under shirt	🟡→🟠→🔴 Alert triggered
Quick grabbing motion	Rapid score increase
Look around nervously	"NERVOUS_LOOKING" detected

Keyboard Controls

Key	Action
Q	Quit application
S	Save screenshot
R	Reset tracking

---

🧠 How It Works

Architecture Overview

┌─────────────────────────────────────────────────────────────────┐
│                        LiftProof.Ai                             │
├─────────────────────────────────────────────────────────────────┤
│                                                                 │
│  ┌──────────┐    ┌──────────────┐    ┌───────────────────────┐ │
│  │  Camera  │───▶│  YOLOv8-Pose │───▶│  Behavioral Analysis  │ │
│  │  Input   │    │  Detection   │    │       Engine          │ │
│  └──────────┘    └──────────────┘    └───────────────────────┘ │
│                         │                        │              │
│                         ▼                        ▼              │
│                  ┌─────────────┐         ┌─────────────┐       │
│                  │  Skeleton   │         │  Suspicion  │       │
│                  │  Tracking   │         │   Scoring   │       │
│                  └─────────────┘         └─────────────┘       │
│                         │                        │              │
│                         └──────────┬─────────────┘              │
│                                    ▼                            │
│                          ┌─────────────────┐                    │
│                          │  Alert System   │                    │
│                          │  & Notification │                    │
│                          └─────────────────┘                    │
│                                                                 │
└─────────────────────────────────────────────────────────────────┘

Detection Pipeline

1. Frame Capture - Video input from webcam or RTSP stream
2. Pose Estimation - YOLOv8-Pose extracts 17 body keypoints
3. Person Tracking - Persistent IDs across frames
4. Behavior Analysis - Analyze keypoint positions and movements
5. Temporal Analysis - Track behavior duration over time
6. Risk Scoring - Calculate 0-100% suspicion score
7. Alert Generation - Trigger alerts when threshold exceeded

Body Keypoints Used

        0: Nose
       / \
      1   2: Eyes
     /     \
    3       4: Ears
     \     /
      5───6: Shoulders
      │   │
      7   8: Elbows
      │   │
      9  10: Wrists ← Primary detection points
      │   │
     11──12: Hips ← Primary detection points
      │   │
     13  14: Knees
      │   │
     15  16: Ankles

---

⚙️ Configuration

Detection Thresholds

Edit these in liftproof_v2.py to tune sensitivity:

self.config = {
    # Distance thresholds (relative to body height)
    'POCKET_THRESHOLD': 0.12,      # Hand near pocket
    'WAIST_THRESHOLD': 0.10,       # Hand at waistband
    'CHEST_THRESHOLD': 0.15,       # Hand under shirt
    
    # Time thresholds (frames at ~30fps)
    'POCKET_FRAMES': 25,           # ~0.8 sec to trigger
    'WAIST_FRAMES': 20,            # ~0.7 sec to trigger
    'CHEST_FRAMES': 25,            # ~0.8 sec to trigger
    
    # Alert thresholds
    'ALERT_THRESHOLD': 70,         # Score to trigger alert
    'SUSPICIOUS_THRESHOLD': 40,    # Score for suspicious
    'WATCHING_THRESHOLD': 20,      # Score to start watching
}

Supported CCTV Brands

Brand	RTSP URL Format
Hikvision	rtsp://{user}:{pass}@{ip}:554/Streaming/Channels/{ch}01
Dahua	rtsp://{user}:{pass}@{ip}:554/cam/realmonitor?channel={ch}
Lorex	rtsp://{user}:{pass}@{ip}:554/ch{ch}/main
Uniview (UNV)	rtsp://{user}:{pass}@{ip}:554/unicast/c{ch}/s0/live
Reolink	rtsp://{user}:{pass}@{ip}:554/h264Preview_{ch}_main

---

📁 Project Structure

LiftProofAI/
├── liftproof_v2.py          # Main detection script (recommended)
├── main.py                   # Multi-camera orchestrator
├── test_webcam.py           # Simple webcam test
├── requirements.txt          # Python dependencies
├── README.md                 # This file
├── LICENSE                   # MIT License
├── .gitignore               # Git ignore rules
│
├── config/
│   └── settings.py          # Configuration settings
│
├── modules/
│   ├── __init__.py
│   ├── detection_engine.py  # Core detection logic
│   ├── rtsp_loader.py       # CCTV stream handler
│   └── notification_manager.py
│
├── models/                   # Trained models (git-ignored)
├── data/evidence/           # Captured alert images
├── logs/                     # Application logs
└── assets/                   # Documentation assets

---

📊 Performance

Benchmarks

Hardware	Resolution	FPS	Persons Tracked
MacBook Air M1	720p	25-30	Up to 5
MacBook Pro M2	1080p	30+	Up to 10
RTX 3060	1080p	45+	Up to 15
RTX 4090	4K	60+	Up to 20+

Model Options

Model	Size	Speed	Accuracy
yolov8n-pose.pt	6 MB	Fastest	Good
yolov8s-pose.pt	23 MB	Fast	Better
yolov8m-pose.pt	52 MB	Medium	Best

---

🔮 Roadmap

• Basic pose detection
• Behavioral analysis engine
• Multi-person tracking
• Webcam support
• CCTV integration
• Mobile app notifications (Firebase)
• SMS alerts (Twilio)
• Web dashboard
• Cloud deployment
• Custom model training pipeline

---

🤝 Contributing

Contributions are welcome! Here's how:

1. Fork the repository
2. Create your feature branch: git checkout -b feature/AmazingFeature
3. Commit your changes: git commit -m 'Add some AmazingFeature'
4. Push to the branch: git push origin feature/AmazingFeature
5. Open a Pull Request

---

📄 License

This project is licensed under the MIT License

---

⚠️ Disclaimer

This software is intended for legitimate security purposes only. Users are responsible for:

• Complying with local privacy laws and regulations
• Obtaining necessary permissions for surveillance
• Proper signage indicating video monitoring
• Ethical use of the technology

---

📞 Support

• Issues: GitHub Issues
• Email: mrifat205@gmail.com

---

🙏 Acknowledgments

• Ultralytics - YOLOv8 framework
• OpenCV - Computer vision library
• Roboflow - Dataset management

---

Built with ❤️ for retail security

⭐ Star this repo if you find it useful!