Real-Time Drone Communication System Using ROS 2 and GStreamer with YOLOv8-Seg for Face Segmentation

This paper presents the design and implementation of a ROS 2-based UAV syste m for real-time video streaming and intelligentground station processing. The proposed architecture integrates a Raspberry Pi 3 onboard computer with a Jetson Orin Nanoground station over a wireless network. Video is captured and encoded using GStreamer on the UAV, streamed over UDP, anddecoded on the ground station for real-time object detection using the YOLOv8-seg model. ROS 2 middleware facilitatessynchronized telemetry and camera communication between the UAV and ground station via DDS topics. The systemdemonstrates low-latency video transmission (∼105 ms), high streaming frame rate (30 FPS), and real-time object detection at28–30 FPS with an average precision of 81.2%. The modularity of ROS 2 enables easy integration of additional perception,control, and autonomous decision-making modules. Experimental results validate the system’s performance for surveillance andinspection tasks, showcasing the potential of open-source middleware and embedded AI for edge-enhanced UAV applications.