The SkyDroid H12Pro uses its own ecosystem of cameras and gimbals, but on my Holybro drone, I use a Luxonis Oak-D Lite camera to infer neural networks. Using the onboard RPI4 computer, I managed to generate a Gstreamer pipeline and a MediaMTX server to create an RTSP video stream. A special twisted pair connects the RPI4 to the Ethernet port of the R12 Pro receiver (on the drone). The video stream is visualized on the H12Pro using the QGC ground control. The video clip provides some details about the setup on my drone. A short demo at the end of the clip shows that the whole concept works well.
holybro drone (2)
This video describes shortly the hardware and software aspects of my self-made tracking drone. The major parts are: a holybro frame, a pixhaxk 6c flight controller, a raspberry pi 4, a oak-d camera and 8 sensors (gps, lidar, compass, ...). The tracking relies on a fine-tuned YOLOv8 object detector and a custom python tracking code. The yaw and the pitch of the drone are derived form the height and center of the detected object. A simple obstacle detection derived from the oak-d stereo camera's, prevents the drone of crashing into obstacles closer than 3m.
