In cities like the Bronx, emergency response time can mean the difference between life and death. I wanted to create something that ensures communities stay safe and that people get immediate help — especially when no one is around to call for it. The idea was to use existing CCTV infrastructure and give it a brain: a real-time AI assistant that can detect danger and alert the right services automatically.
🛠️ How I Built It I combined several technologies to bring SafeWatch AI to life:
MediaPipe and OpenCV for real-time video analysis.
Custom AI logic to detect various emergencies such as:
Fires
People fainting
Acts of violence or suspicious behavior
Text-to-Speech (TTS) systems to generate human-like alerts.
Automated calling scripts to contact emergency services (e.g., 911) based on AI predictions.
A simple configuration system to define which camera is monitoring which area, allowing tailored alert messages like:
“This is an automated message from the AI system monitoring 161st Street, Bronx. A fire has been detected. Please respond immediately.”
The script runs continuously, processes video feeds live, and is capable of handling multiple incidents simultaneously.
📚 What I Learned How to integrate AI with real-time video analysis for meaningful use cases.
The power of combining machine learning with old infrastructure (like CCTVs) to create something truly useful.
How to automate emergency communication in a real-world context.
⚠️ Challenges Faced Making AI reliable enough to detect complex scenarios like fainting or aggressive behavior.
Ensuring false positives are minimized to avoid unnecessary emergency calls.
Setting up voice call automation without relying on paid APIs like Twilio.
Keeping performance high enough for real-time operation on potentially limited hardware.
💡 Future Improvements Add face recognition (if ethical/legal) for identifying victims.
Use anomaly detection for subtle threats or early signs of danger.
Connect with local community apps for non-critical alerts (e.g., a missing person or minor injury).
Launch this idea based project to a real world code.
Log in or sign up for Devpost to join the conversation.