Road Hazard Detection & Real-Time Alerts

Inspiration

Everyday countless, accidents occurs due to road hazards like potholes,animal crossing the road etc .Especially in developing countries like India,Bangladesh etc

What it does

The aim of the project was to create an ai powered system that can detect these hazards in real time using a simple dashcam feed,helping drivers to stay alert and react faster.

How we built it

Frontend (React.js)

Displays the live video feed from an IP Webcam (mobile camera).
Shows hazard alerts dynamically without requiring a page refresh.
Communicates with the Go backend using gRPC.

Backend (Golang)

Handles live frame requests and routes them to the Python model. 2.Manages real-time data flow between the frontend and detection module. 3.Designed for low latency and offline functionality over local Wi-Fi. 4.We used the Ollama LLaVA model for hazard detection, but we can also create a custom Python model server using the YOLO dataset. We can also use Gemini because the LLaVA model requires a GPU to run efficiently.

Challenges we ran into

1 . Running the system entirely offline meant no cloud inference or external APIs could be used.

This required careful resource management, especially for real-time image processing and alert display. ## Accomplishments that we're proud of The hazard detection part can completely run offline reducing the latency issue , using gRPC further reduced it. ## What we learned 1.. The importance of optimizing ML models for edge devices like Raspberry Pi.
Handling network latency and synchronizing video streams efficiently.
Designing a system that can run completely offline using local resources.