video-logo-tracker

⚙️ Overview

The system consists of two main components:

API

A FastAPI application deployed on Google Cloud Run. It allows users to:

• Upload a video and a logo image.
• Trigger cloud function that processes the video to detect occurrences of the logo.

Worker

A Google Cloud Function responsible for the actual video processing. It uses the SIFT (Scale-Invariant Feature Transform) algorithm to:

• Detect the logo in individual frames.

• Generate an output video with detected logo regions highlighted.

🚀 Live Demo

🔗 Visit the App: https://neurons-code-challenge.lovable.app/

🔗 Read the API docs: https://video-logo-api-606507129150.europe-west1.run.app/docs

• Worker auto scaling for concurrent requests, cost effective (min instance 0)
• Decoupled Architecture (separate long processing workflow from API)

Future improvements

• Computational Resources: speed, memory and CPU
• Job queuing, retry
• SIFT might not work in 3D, like logo on curved object
• Limitations in the presence of complex background textures and occlusions

Suggestions

• Stream read and write video
• Skip frames
• Lower resolution
• Use Cloud Task to queue jobs and retry