S.C.O.P.E. is an AI-based multi-camera surveillance and monitoring system designed for real-time person detection, tracking, cross-camera re-identification, and intelligent alert generation.
The project focuses on transforming traditional CCTV systems from passive recording tools into intelligent real-time surveillance systems capable of assisting security personnel proactively.
Inspired by a real theft incident in Chenab Hostel at IIT Ropar, the system was developed to address a major limitation of existing surveillance infrastructure: the absence of intelligent correlation and continuous tracking across multiple cameras.
- Real-time person detection using YOLOv11m
- Cross-camera identity tracking and re-identification
- Multi-stage cascade tracking pipeline
- Kalman Filter based motion prediction
- Event-driven real-time alert generation
- Real-time dashboard synchronization using Socket.IO
- Suspicious activity and dwell-time monitoring
- RTSP and multi-camera stream support
- Centralized monitoring dashboard
- Modular and scalable architecture
- Python
- PyTorch
- OpenCV
- Ultralytics YOLO
- TorchReID (OSNet)
- NumPy
- SciPy
- Flask
- Socket.IO
- React
- TypeScript
- Vite
- Tailwind CSS
- Socket.IO Client
Camera Feed
↓
YOLOv11m Person Detection
↓
Feature Encoding (OSNet)
↓
Kalman Filter + Multi-Stage Tracking
↓
Cross-Camera Re-Identification
↓
Event Generation
↓
Real-Time Dashboard + Alerts
The system uses YOLOv11m for real-time person detection. A dual-confidence strategy is employed to preserve identities during temporary occlusions and low-confidence detection periods.
A Kalman Filter predicts object positions across frames, helping maintain tracking continuity during temporary detection loss.
The tracking pipeline uses a multi-stage cascade assignment strategy to prevent unstable detections from disrupting existing identities.
This improves:
- Identity stability
- Occlusion handling
- Track continuity
- Re-identification consistency
Each detected individual is encoded into a 512-dimensional embedding vector using OSNet.
Cross-camera matching is performed through:
- Vectorized similarity search
- Margin-based filtering
- Transit-time validation
This allows the system to maintain a global identity across multiple camera feeds.
Instead of sending raw tracking data continuously, the system emits meaningful events such as:
- TRACK_ACTIVATED
- PERSON_REIDENTIFIED
- ALERT_TRIGGERED
- TRACK_TERMINATED
Each event contains:
- Identity information
- Camera source
- Timestamp
- Contextual metadata
The centralized dashboard enables security personnel to:
- Monitor live feeds
- Track individuals across cameras
- Visualize movement paths
- Access event logs
- Receive real-time alerts
The frontend remains synchronized using persistent Socket.IO connections.
A lightweight L1 distance filter eliminates spatially unlikely track-detection pairs before expensive computations.
Each identity maintains a bounded set of high-confidence embeddings using a min-heap structure to improve robustness against noisy samples.
Insertion Complexity = O(log k)
Identity matching is executed using vectorized matrix operations instead of sequential loops, significantly improving runtime performance.
A deque and hash-map based dual buffer structure enables:
- Constant-time insertion
- Fast retrieval
- Efficient identity revival
- Ordered eviction
O(1) Operations
The assignment problem is decomposed into multiple stages to prioritize stable tracks and reduce identity fragmentation.
O(max(T, D)^3)
git clone https://github.com/248Vansh/SCOPE.git
cd SCOPEpython -m venv venvvenv\Scripts\activatesource venv/bin/activatepip install -r requirements.txtpython -m core.orchestratornpm install
npm run devConfiguration files:
config/config.yaml
config/cameras.yaml
These files manage:
- Camera streams
- Detection thresholds
- Tracking parameters
- Re-identification settings
- Runtime behavior
- Campus Security
- Hostel Surveillance
- Smart Public Monitoring
- Railway Stations
- Institutional Safety
- Smart Surveillance Networks
The system may face reduced performance under:
- Poor lighting conditions
- Heavy crowd density
- Severe occlusions
- Similar appearances among individuals
- Network instability
- Power interruptions
The system is designed with awareness of privacy and ethical concerns associated with surveillance technologies.
Key considerations include:
- Secure data handling
- Controlled access mechanisms
- Transparent usage policies
- Responsible deployment practices
Balancing security and privacy remains a critical objective.
- Face recognition integration
- Edge-device deployment
- Distributed multi-node processing
- Behavioral anomaly detection
- Privacy-preserving AI mechanisms
- Scalable city-wide surveillance integration
The project was inspired by a real theft incident in Chenab Hostel at IIT Ropar, where existing CCTV infrastructure failed to provide actionable intelligence despite continuous recording.
While cameras captured footage, there was:
- No cross-camera identity continuity
- No real-time alert generation
- No automated tracking
- No intelligent event correlation
S.C.O.P.E. was developed to bridge this gap by enabling intelligent, real-time surveillance capable of assisting security personnel proactively rather than reactively.
This project is intended for academic and research purposes only.