An AI-Driven Real-Time Surveillance Framework for Proacti ve Campus Security
Click the image below to watch the demo video:
- Overview
- Key Features
- System Architecture
- Technology Stack
- Installation
- Configuration
- Usage Guide
- API Endpoints
- Project Structure
- Database Schema
- Troubleshooting
- Performance Benchmarks
- Contributing
- License
- Author
CampusGuard AI is an intelligent, real-time surveillance system designed specifically for campus security. Leveraging state-of-the-art YOLOv11 object detection models, it provides proactive monitoring, anomaly detection, and incident management to ensure campus safety.
Unlike traditional CCTV systems that only record footage, CampusGuard AI understands what it sees—detecting potential threats, unusual behaviors, and security incidents in real-time. The system can identify fights, sleeping individuals, suspicious loitering, and mobile phone usage, making it an invaluable tool for modern campus security.
- Proactive Security: Detect incidents before they escalate
- Multi-Model Intelligence: Specialized models for different threat types
- Real-Time Processing: Instant alerts and notifications
- Cost-Effective: Runs on existing hardware with optional GPU acceleration
- User-Friendly: Intuitive web interface for security personnel
| Model | Detection Capability | Confidence Threshold | Use Case |
|---|---|---|---|
| Fight Detection | Physical altercations, aggressive movements | 70% | Crowded areas, parking lots, sports complexes |
| Sleep Detection | Inactive/lying persons, drowsiness | 25% | Classrooms, libraries, lecture halls |
| Suspicious Behavior | Loitering (4+ minutes), unusual patterns | 90% | Restricted areas, after-hours zones |
| Normal Detection | Standard person/object tracking | 50% | General surveillance across campus |
| Phone Detection | Mobile device usage | 75% | Exam halls, restricted zones |
Fight
- Webcam (built-in laptop cameras) - Perfect for testing and small-scale deployment
- External USB cameras - Automatic device detection and indexing
- IP cameras via RTSP/HTTP streams - Support for existing CCTV infrastructure
- Live streaming with real-time frame processing
- Frame caching for efficient display and reduced bandwidth
- Configurable confidence thresholds per model and per camera
- Delay-based incident creation prevents false positives and notification spam
- Automatic evidence capture with bounding boxes and timestamps
- Severity classification (Low, Medium, High, Critical) for prioritization
- Complete incident lifecycle tracking from detection to resolution
- Base64 encoded evidence storage for easy display and sharing
- Video evidence paths for larger incident files
- Multi-camera monitoring grid with individual controls
- Per-camera model toggles - Enable/disable models on the fly
- Live FPS and detection statistics for performance monitoring
- Anomaly alerts with browser notifications and sound alerts
- Incident timeline with evidence preview and quick actions
- System status indicators (GPU availability, model health)
| Role | Permissions |
|---|---|
| Admin | Full system control, user management, camera management, database operations |
| Security | Incident management, monitoring controls, report viewing |
| Faculty/Student | Limited access, incident reporting, basic dashboard view |
- CUDA/GPU acceleration for 5-10x faster inference
- Multi-threaded stream processing for concurrent camera handling
- Queue-based frame management prevents bottlenecks
- JPEG compression with configurable quality balances speed and clarity
- Optimized OpenCV backends (DSHOW, MSMF, VFW) for Windows compatibility
- Lazy loading of models to reduce startup time