Inspiration
What it does
WatchOut is an AI-powered crime detection system that connects with existing CCTV or phone cameras and sends instant alerts when violent behaviors are detected.
WebSocket live video streaming Real-time behavior recognition (fights, assaults, etc.) Push alerts to verified community members FastAPI backend + secure communication Mobile + web live feed viewing
At its core, our model predicts behavior on a continuous stream of video frames, learning how movement changes over time to detect threats as they unfold.
How we built it
AI Model: CNN-based violence classifier enhanced with temporal motion analysis
Backend: FastAPI server with WebSocket streaming for low-latency inference
Frontend: Made with React and Typescript
Mobile App: Expo + React Native client for real-time frame streaming & overlays
Optimizations: Adaptive backpressure, lightweight JPEG compression, GPU acceleration where available
Challenges we ran into
OAuth token flow over continuous streaming sessions
Keeping persistent WebSocket connections stable
Designing an efficient real-time video streaming protocol
Containerizing our model for deployment
Debugging asynchronous callbacks between frontend and backend
Accomplishments that we're proud of
Build a clean and intuitive UI/UX that empowers users rather than overwhelms them
Successfully train and integrate a real-time crime detection model
Deliver a full-stack system spanning mobile, web, and secure backend infrastructure
What we learned
Stream and process video in real time
Apply temporal AI models for behavior recognition
Build secure, full-stack systems from scratch
Balance performance, accuracy, and ethical design
What's next for Watchout: Crime Detection Program
More threat categories
Stronger alert verification
Edge-device deployment
Smart CCTV integration
Scalable + privacy-first architecture
Log in or sign up for Devpost to join the conversation.