SafeCity AI

SafeCity AI uses real-time computer vision to detect accidents and violence from CCTV feeds, instantly alerting authorities to enable faster emergency response and safer communities.

Comment

Inspiration

Road accidents and violent incidents often go unnoticed until someone reports them, delaying emergency response and putting lives at risk. We wanted to build an AI system that transforms traditional CCTV cameras into intelligent observers capable of detecting critical events in real time. Our goal was to leverage computer vision to improve public safety and support faster, data-driven emergency response.

What it does

SafeCity AI is an AI-powered surveillance platform that analyzes live video streams to detect road accidents and violent incidents in real time. When an incident is identified, the system generates instant alerts, highlights the event on a monitoring dashboard, and provides visual evidence to help authorities respond quickly. Instead of passively recording footage, SafeCity AI actively monitors public spaces and delivers actionable insights for smarter and safer cities.

How we built it

We developed SafeCity AI using a modern AI and web architecture:

  • YOLOv8 for real-time object detection
  • OpenCV for video processing and frame management
  • PyTorch for deep learning inference
  • FastAPI for serving AI models through REST APIs
  • React.js for building the interactive monitoring dashboard
  • Docker for containerized deployment and scalability

The system processes live video streams, performs AI inference, detects incidents, and sends results to the dashboard with minimal latency.

Challenges we ran into

One of our biggest challenges was achieving reliable real-time detection while maintaining low latency. Balancing inference speed with detection accuracy required careful model optimization and efficient frame processing. We also faced challenges in handling different lighting conditions, camera angles, and crowded scenes, ensuring the system remained responsive across diverse environments.

Accomplishments that we're proud of

  • Built a complete end-to-end AI surveillance platform.
  • Achieved real-time accident and violence detection on live video streams.
  • Designed a scalable backend and interactive dashboard for monitoring incidents.
  • Combined AI, backend engineering, and frontend development into a production-style application.
  • Demonstrated how computer vision can be applied to solve real-world public safety challenges.

What we learned

This project strengthened our understanding of deploying computer vision models in real-time applications. We learned how to optimize AI inference pipelines, build scalable APIs, integrate AI with modern web technologies, and design systems that are both technically robust and practical for real-world deployment.

What's next for SafeCity AI

Our roadmap includes:

  • Multi-camera monitoring and centralized management.
  • Automatic emergency notifications to authorities.
  • GPS mapping and incident heatmaps.
  • Edge deployment using TensorRT and NVIDIA Jetson devices.
  • Cloud-based analytics and historical reporting.
  • Expanded detection capabilities for fire, weapons, crowd anomalies, and other public safety events.
  • AI-powered incident summaries and predictive safety analytics for smart cities.

Built With

Share this project:

Updates

Submission history