Advanced Multi-Camera Surveillance with Access Control

This project is a real-time, multi-camera surveillance system that detects, tracks, and re-identifies people across different video streams. It includes Cross-Camera Re-Identification, Zone-Based Access Control, and Immutable Audit Logs. The system leverages state-of-the-art deep learning models for high accuracy and provides a robust alerting system for both unknown individuals and unauthorized zone entries.

🚀 Key Features

• Multi-Camera Support: Processes multiple video streams (from files or live cameras) concurrently using threading.
• Cross-Camera Re-Identification:
  • Global Tracking: Assigns a unique, consistent ID to a person as they move between different cameras.
  • Transition Logging: Automatically detects and logs when a person exits one camera view and enters another.
• Immutable Audit Logs:
  • Tamper-Evident: Uses SHA-256 hash chaining to link all log entries.
  • Secure Signing: All logs and file artifacts (snapshots) are signed with an HMAC using a secure key.
  • Comprehensive Logging: Tracks all alerts, administrative actions (Enroll/Edit/Delete), and data exports.
• Person Detection & Tracking:
  • YOLO: Fast and accurate person detection.
  • DeepSORT: Robust local tracking within single camera views.
  • OSNet (Re-ID): Extracts deep feature vectors to recognize individuals across different times and cameras.
• Dual Alert System:
  • Unknown Person Alert: Triggers if an unknown person remains in view for too long.
  • Unauthorized Access Alert: Triggers if a known person enters a restricted camera zone.
• Centralized Dashboard: Displays all camera feeds in a single grid view.

---

🛠️ How It Works (System Architecture)

1. Frame Capture: The system reads frames from multiple video sources.
2. Detection & Local Tracking: YOLO detects people, and DeepSORT assigns local track IDs.
3. Cross-Camera Integration:
   • The GlobalTrackManager receives features from all cameras.
   • It matches new tracks against existing global identities using Re-ID embedding similarity and spatio-temporal constraints.
   • If a match is found, the person is assigned their existing Global ID.
4. Identification: Features are compared against the MongoDB people collection.
5. Audit Logging:
   • Every alert, access violation, or system action is sent to the AuditLogger.
   • The logger computes a hash of the event data + the previous log's hash, then signs it with a secret key.
   • The entry is stored in the audit_ledger database.
6. Visualization: Frames are annotated with Global IDs and names, then displayed on the dashboard.

---

⚙️ Tech Stack

• AI / ML: PyTorch, ultralytics (YOLO), torchreid (OSNet), deep-sort-realtime
• Database: MongoDB (Stores People, Access Rules, History, and the Immutable Audit Ledger)
• Security: HMAC, SHA-256 (for Audit Logs)
• Core: OpenCV-Python, NumPy, SciPy, Flask (Web Interface)

---

🏁 Getting Started

Prerequisites

• Python 3.8+
• MongoDB installed and running on localhost:27017.
• NVIDIA GPU (Recommended).

1. Installation

git clone https://github.com/adityapatil37/PIDS
cd PIDS

# Create and activate venv
python -m venv venv
# Windows: venv\Scripts\activate
# Linux/Mac: source venv/bin/activate

# Install dependencies (ensure PyTorch matches your CUDA version)
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118
pip install -r requirements.txt

2. Database Setup

Ensure MongoDB is running. The application will automatically create:

• person_reid DB: For users, access rules, and history.
• audit_ledger DB: For the immutable audit chain.

3. Run the Application

python ids.py

• Dashboard: A window showing live camera feeds with tracking.
• Web Interface: Access http://localhost:5000 (if main.py is running) for enrollment and history.

4. Admin Web Interface

Run the Flask app to manage users and view logs:

python main.py

• Enrollment: Upload images to register new people.
• History: View attendance logs.
• Access Control: Configure which cameras specific people can access.
• Alerts: View unauthorized access incidents.

---

🛡️ Security Features

Audit Log Verification

To verify the integrity of the audit logs and ensure no tampering has occurred:

python test_audit_log.py

This script will:

1. Traverse the entire hash chain in MongoDB.
2. Re-compute hashes and HMAC signatures for every entry.
3. Report any broken links or invalid signatures.

---

📁 Project Structure

.
├── ids.py                  # Main surveillance system
├── main.py                 # Flask Web Admin Interface
├── global_tracker.py       # Cross-Camera Tracking Logic
├── audit_manager.py        # Immutable Audit Log Implementation
├── alert_snapshots/        # Signed images of alerts
└── static/thumbnails/      # Signed thumbnails of detections

---

📜 License & Acknowledgements

Copyright © 2025 Aditya Patil. All rights reserved.