Build Your Own Selfie Drone With Computer Vision

In late 2013 and early 2014, in the heady days of the drone revolution, there was one killer app — the selfie drone. Selfie sticks themselves had already become a joke, but a selfie drone injected a breath of fresh air into the world of tech. Fidget spinners had yet to be invented, so this is really all we had. It wasn’t quite time for the age of the selfie drone, though, and the Lily camera drone — in spite of $40 Million in preorders — became the subject of lawsuits, and not fines from the FAA.

Technology marches ever forward, and now you can build your own selfie drone. That’s exactly what [geaxgx] did, although this build uses a an off-the-shelf drone with custom software instead of building everything from scratch.

For hardware, this is a Ryze Tello, a small, $100 quadcopter with a front-facing camera. With the right libraries, you can stream images to a computer and send flight commands back to the drone. Yes, all the processing for the selfie drone happens on a non-flying computer, because computer vision takes processing power and battery life.

The software comes from CMU’s OpenPose library, a real-time solution for detecting a body, face, or hands. With this, [geaxgx] was able to hover the drone and keep his face in the middle of the camera’s frame. While there’s no movement of the drone involved — the drone is just hovering and rotating to the left and right — it is a flying selfie stick without the stick. You can check out the video below and check out all the code on [geaxgx]’s GitHub here.