Real-Time Collision Detection with YOLOv8, MiDaS & IoU Tracking

The Problem: 2D Detection Isn't Enough

Traditional object detection gives you bounding boxes — but not depth. For collision detection, knowing how far an object is matters as much as detecting it. That's why CollisionGuard AI combines YOLOv8 with MiDaS monocular depth estimation.

The Risk Scoring Engine

I built a multi-factor 0–100 risk scoring engine using IoU-based motion tracking. It factors in object class, relative velocity, depth, and trajectory — giving a quantified collision risk in real time.

WebSocket Streaming with FastAPI

The backend uses FastAPI with WebSocket streaming to push live predictions to the React/TypeScript frontend, achieving sub-100ms latency for real-time use cases.