Backend & Data Engineer

← Back to home

Computer-vision speed detection API

Problem

Vehicle speed from 9 cameras had to be served as a production API.

Approach

A FastAPI service estimates vehicle speed from those cameras with YOLO and returns the speed. The service ships as Docker images, with unittest suites.

Outcome

About 15k requests over 141 days (about 106 a day), 99.91% success, and zero 5xx errors.

Amber flowchart on a black field: request, confirm caller, then take a camera clip and plate in parallel with load road segment, detect vehicles, then read plates and estimate speed, mark detection, store decision, and return result.

Skills used

  • FastAPI
  • YOLO
  • PyTorch
  • Docker
  • PostgreSQL
  • MongoDB