Spring 2024 · AI Trainer
Drone Tracking
A computer-vision model that finds and labels people in live drone video, streamed through a server to a Flutter app for viewing and archiving.
- Role
- AI Trainer
- When
- Spring 2024
- Discipline
- AI & ML
Built with
Overview
The goal was to spot people from the air, live. We streamed video from a drone to a server, ran it through a model that annotates every person it finds, and pushed the annotated stream to an app.
Pipeline
- Collect. We recorded 30 minutes of video streamed from the drone.
- Annotate. I labelled that footage by hand so the model had ground truth to learn from.
- Train. The model learned what a person looks like from above and how to mark one.
- Serve. The drone streams live video to our server, where the model annotates each frame.
- View. The server streams the annotated video to a Flutter app the team built, where it can be watched and archived.
Results
Two clips annotated by the model:
