sex offender identification demonstration
main security dashboard
ai threat evaluation demonstration
motion tracking demonstration
sex offender doxxing demonstration

About Sentri

Inspiration

The inspiration for Sentri came from the need for comprehensive, intelligent surveillance systems that can protect communities by identifying potential threats while maintaining privacy for family members. We recognized that traditional security cameras only provide passive monitoring without intelligent analysis. The goal was to create a system that could automatically detect and alert on registered sex offenders in real-time, while distinguishing between threats and family members, making neighborhoods safer through proactive AI-powered surveillance.

What it does

Sentri is a distributed multi-camera surveillance system that combines computer vision, face recognition, and AI analysis to provide intelligent security monitoring. The system automatically detects faces in real-time video streams, cross-references them against a sex offender database, and generates AI-powered activity reports. Key features include:

Real-time Face Detection: Automatically identifies faces in video streams from multiple cameras
Sex Offender Recognition: Cross-references detected faces against a comprehensive sex offender database with 17+ registered offenders
Family Member Management: Recognizes and protects family members from false alerts
AI-Powered Analysis: Uses Google Gemini AI to generate natural language reports of activities
Distributed Architecture: Supports multiple camera clients across different machines
Web Interface: Modern dashboard for monitoring, configuration, and alert management
Vector Database Search: Semantic search capabilities to find similar events using natural language

How we built it

Sentri was built using a comprehensive tech stack combining computer vision, machine learning, and web technologies:

Backend Technologies:

YOLOv8: Real-time object detection and tracking
OpenCV: Computer vision and video processing
Face Recognition: Facial feature extraction and matching using dlib
Google Gemini AI: Natural language analysis and report generation
ChromaDB: Vector database for semantic search
Flask: Web application framework with WebSocket support

Frontend Technologies:

HTML5/CSS3: Modern responsive web interface
JavaScript: Real-time updates and camera controls
WebSocket: Live video streaming and notifications

Infrastructure:

Distributed Architecture: Multiple camera clients connected to central server
Selenium Web Scraping: Automated data collection from sex offender registries
SQLite Database: Local storage for face embeddings and offender data
REST API: Comprehensive API for system control and data access

Data Collection:

Automated web scraping of New Jersey State Police sex offender registry
Successfully extracted 17 offenders within 5-mile radius with complete profiles
Downloaded 10 offender mugshots for facial recognition database

Challenges we ran into

Technical Challenges:

Anti-Bot Protection: The sex offender registry used DataDome protection that blocked traditional web scrapers. We overcame this by implementing Selenium with proper session management and respectful rate limiting.
Face Recognition Accuracy: Balancing false positives and false negatives required extensive tuning of confidence thresholds and implementing multiple validation layers.
Real-time Performance: Processing multiple camera streams simultaneously while maintaining real-time performance required optimization of the video processing pipeline and implementing efficient frame buffering.
Distributed System Coordination: Managing multiple camera clients with different hardware capabilities and network conditions required robust error handling and automatic reconnection mechanisms.

Data Challenges:

Limited Image Quality: Some sex offender mugshots were low quality or unavailable, requiring fallback strategies and manual verification.
Privacy Compliance: Ensuring the system respects privacy while providing security required careful implementation of family member recognition and data retention policies.

Integration Challenges:

AI Cost Management: Gemini API costs needed careful optimization through prompt engineering and selective analysis to keep operational costs under $0.0002 per analysis.
Cross-Platform Compatibility: Ensuring the system works across different operating systems and hardware configurations required extensive testing and platform-specific optimizations.

Accomplishments that we're proud of

Complete End-to-End System: Successfully built a production-ready surveillance system with web interface, distributed architecture, and AI integration
High Accuracy Face Recognition: Achieved reliable face detection and matching with proper family member protection
Cost-Effective AI Integration: Implemented intelligent AI analysis at under $0.0002 per report, making it affordable for continuous operation
Comprehensive Data Collection: Successfully scraped and processed 17 sex offender profiles with 10 high-quality images
Real-time Performance: Achieved real-time processing of multiple camera streams with live WebSocket updates
Modern Web Interface: Created an intuitive, responsive dashboard with real-time notifications and comprehensive controls
Robust Architecture: Built a distributed system that can scale across multiple machines with automatic error recovery
Semantic Search: Implemented vector database search allowing natural language queries like "find suspicious activities"

What we learned

Technical Learnings:

Computer Vision Integration: Gained deep experience in combining multiple computer vision technologies (YOLO, OpenCV, face recognition) into a cohesive system
Distributed Systems: Learned how to design and implement distributed architectures with proper client management, load balancing, and fault tolerance
AI Integration: Mastered the art of integrating large language models into real-time systems while managing costs and performance
Web Scraping: Developed expertise in bypassing modern anti-bot protections using Selenium and proper session management

Project Management:

Modular Development: Learned the importance of building modular, reusable components that can be easily integrated and maintained
User Experience: Discovered how critical intuitive interfaces are for complex technical systems
Performance Optimization: Gained insights into optimizing real-time video processing pipelines for production use

Ethical Considerations:

Privacy vs Security: Learned to balance security needs with privacy protection, implementing family member recognition to prevent false alarms
Responsible AI: Understood the importance of transparency and user control in AI-powered surveillance systems

What's next for Sentri

Short-term Enhancements:

Mobile App: Develop iOS and Android applications for remote monitoring and push notifications
Advanced Analytics: Implement behavior pattern recognition and predictive threat assessment
Cloud Integration: Add cloud storage for event history and backup capabilities
Enhanced AI: Integrate more sophisticated AI models for better activity classification and threat assessment

Medium-term Goals:

Multi-Region Support: Expand sex offender database to cover multiple states and regions
Integration APIs: Create APIs for integration with existing security systems and smart home devices
Advanced Reporting: Implement automated incident reports with detailed timelines and evidence compilation
Machine Learning Improvements: Train custom models on surveillance data to improve detection accuracy

Long-term Vision:

Community Network: Create a network of Sentri systems that can share threat intelligence across neighborhoods
Predictive Security: Develop predictive algorithms that can identify potential security threats before they occur
Compliance Framework: Build tools to help communities ensure compliance with local surveillance laws and privacy regulations
Open Source Initiative: Release core components as open source to enable community-driven development and transparency

Technical Roadmap:

Edge Computing: Implement edge processing capabilities to reduce bandwidth and improve response times
Blockchain Integration: Use blockchain for tamper-proof event logging and audit trails
Advanced Encryption: Implement end-to-end encryption for all data transmission and storage
Scalability Improvements: Design for deployment across large residential communities and commercial properties