After taking a brief pause from this project during finals week, I’m excited to share that I finally got real-time drone tracking working.

The system can now detect my drone from a live camera feed using a custom-trained YOLOv8 model and draw reliable bounding boxes around it. From there, I calculate the center of the drone’s bounding box, compare it to the center of the camera frame, and use that offset to track where the drone is moving. I also started mapping that position onto a 3D graph using OpenCV, with only one camera instead of a multi-camera setup.

This took quite a bit of math and debugging. I had to work through monocular camera geometry, pixel offsets, scaling, and OpenCV visualization before the tracking started making sense. I also trained and tested my own drone detection model, then built the pipeline so the model can run live, pick the highest-confidence drone detection, and output useful position/error data in real time.

The next step is connecting this tracking system to the physical turret. The code now calculates horizontal and vertical error values from the drone’s position in the frame, which can be sent to an Arduino to control pan and tilt movement.

Currently I’m working on “hacking” into this small drone I own, and doing some resoldering and rewiring so that this store-bought drone can be controlled digitally though my Arduino instead of through a human with its remote. Then using this digital control system, I can then intercept the larger flying drone and take it down.