Proof of concept Jonathan Elliott·2239 AI Fundamentals ·View code on GitHub ↗

Drone Vision Safety Check

A proof-of-concept image model, trained in Google Teachable Machine, that tells a drone whether what its camera sees is safe or dangerous.

Drone camera

Start the camera or upload a photo to test the model.
CAM OFF

Model decision

WAITING
Loading model…

Sample photos

Click a photo to see what the drone decides.

Note: none of these sample photos were used to train the model. They are new images it has never seen, so they show how well the model works on real-world pictures.

Loading samples…

How it works

Each camera frame is passed to an image classifier trained on two classes:

Limitation: the model classifies the whole frame. It doesn't locate the obstacle or measure its distance, which a real drone would need from object detection and depth sensors.