Air Patrol | Devpost

Inspiration

When we played around with the drone for the first time, we immediately knew we wanted to build something with it. We wanted to incorporate two main features: 1) being able to see places/angles that would normally be difficult/inconvenient and 2) Automation with computer vision. After brainstorming problems that we could solve, we landed on Police vest badges and decided to make a drone supplement to improve this system.

What it does

The underlying technology is Computer Vision to recognize people and stay within a certain distance from them. Our application is to help provide a better view when compared to the traditional Police vest badges. It is also safer and enables a variety of angles.

How we built it

We used OpenCV for the Computer Vision aspect, along with a pre-trained XML recognition model. All the back-end was done in Python (calculating distances and moving the drone). The Tello library was used for all drone commands (starting up and moving)

Challenges we ran into

Connecting the Camera to OpenCV and getting a live feed was extremely difficult to figure out. We had a lot of problems with OS compatibility (Tello is built for Ubuntu not Arch Linux :( ), so we weren't able to install many of the recommended packages.

Accomplishments that we're proud of

Finally figuring out how to use OpenCV with the Tello library. This challenge took up majority of our hacking time and we felt extremely accomplished afterwards.

What we learned

Our primary learnings were how to use the OpenCV and Tello libraries. It was the first time anyone in our group had done a drone hack, so there was a lot to learn from it!

What's next for Air Patrol

Training the model to target specific people, not necessarily from their faces (i.e. behind their head). Improving the accuracy of our computer vision model. Upgrading the drones used in the project (increasing stability, battery life, etc.)