A-eye

Home Page
Live Threat Detection
Live Threat Detection

Inspiration

With rising concerns around public safety, existing CCTV systems remain largely passive — they record incidents but cannot prevent them. We wanted to build something that could detect danger before it strikes. The inspiration behind A-eye came from real-world tragedies that might have been avoided with faster response times. Our goal was to create an AI-powered system that transforms ordinary cameras into proactive guardians capable of recognizing threats in real time and saving lives.

What it does

A-eye is a real-time threat detection and alerting system powered by computer vision. It continuously analyzes live CCTV footage to identify weapons, violent actions, or suspicious behavior. Once a threat is detected, the system instantly alerts administrators and first responders through a connected dashboard, providing camera location, time, and visual context. A-eye not only warns but also helps coordinate an immediate and targeted response, turning traditional surveillance into active prevention

How we built it

We developed A-eye using a combination of edge computing and AI:

Computer Vision: Built on the YOLOv8 object detection model for fast, accurate recognition of weapons and aggressive behavior.
Hardware Integration: Implemented on Raspberry Pi 4 with the IMX519 camera module for edge-based real-time processing.
Backend: Designed with FastAPI for alert handling, camera management, and RESTful APIs.
Frontend: Created a clean and responsive React dashboard for monitoring alerts, live feeds, and system logs.
Authentication: Integrated Auth0 for secure multi-role access control between administrators and responders.
Alerts: Configured automated notifications via Twilio (SMS) and Discord webhooks for instant communication.

Challenges we ran into

Hardware Optimization: Running deep learning inference on Raspberry Pi with limited resources required careful model tuning and frame-rate balancing.
False Positives: Ensuring accuracy without triggering unnecessary alerts meant fine-tuning detection thresholds and confidence levels.
Integration Issues: Synchronizing video feeds, backend APIs, and real-time alert systems across devices introduced latency and synchronization hurdles.
Privacy Considerations: We designed the system to automatically blur faces and avoid storing sensitive footage, ensuring responsible AI deployment.

Accomplishments that we're proud of

Built a fully functional AI-powered threat detection system that operates in real time on low-power hardware.
Integrated live alerts and monitoring through a centralized dashboard.
Successfully deployed a privacy-focused, scalable design that could easily extend to schools, offices, and public venues.
Created a reliable, automated pipeline from detection to notification in under a second.

What we learned

Optimizing computer vision models for edge devices requires both software and hardware-level efficiency.
Real-time systems demand careful design around latency, concurrency, and error handling.
Security and ethics are as critical as accuracy — AI must protect without invading privacy.
Collaboration between AI, hardware, and cloud integration teams is essential for scalable, real-world safety applications.

What's next for A-eye

Expanding detection capabilities beyond weapons to include fight recognition and abnormal movement analysis.
Implementing cloud-based dashboards with analytics, heatmaps, and predictive threat modeling.
Integrating with city surveillance networks and emergency dispatch systems for automated response coordination.
Refining model accuracy through continuous learning from verified incident data.
Exploring multi-camera tracking and behavior analytics to provide deeper situational awareness.