Drone-Based Object Tracking and Track Data Transmission to a Command and Control System

Sammanfattning

This thesis presents the design, implementation, and evaluation of a Unmanned Aerial Vehicle (UAV)-based electro-optical tracking system that delivers track-level information to a ground-based Command and Control (C2) system. An onboard processing pipeline running on an NVIDIA Jetson platform performs real-time object detection using TensorRT Inference Optimizer (TensorRT)-accelerated You Only Look Once (YOLO) models and the system performs multi-object tracking in the image plane using a linear Kalman Filter. To estimate the target’s position in 3D space, an Extended Kalman Filter (EKF) is employed for Target Motion Analysis (TMA), converting bearing-only measurements into Cartesian coordinates (x, y, z). Detected targets are converted from image coordinates to bearing measurements (azimuth and elevation) and compensated for UAV platform attitude using rotation matrices derived from Micro Air Vehicle Link (MAVLink) telemetry. Track data are transmitted to the ground using standard MAVLink messages, where a router distributes the traffic to QGroundControl (QGC) and to an adapter that translates the data to protobuf format for forwarding to the Track Data Fusion Engine (TDFE) C2 system. A reverse communication path allows the C2 system to return correlated target cues to the UAV, enabling onboard target selection and follow behavior. Experiments demonstrate that the pipeline achieves real-time operation at approximately 8-10 fps with an end-to-end latency of 100-125 ms, and that the produced track data are suitable for ingestion by downstream fusion systems. Limitations include bearing-only observability, monocular range uncertainty, and communication-link dependencies. The work contributes a complete prototype architecture for UAV-to-C2 track delivery, including message format selection, protocol adaptation, and bidirectional integration.