Lena Brüggemann
FPGA-based Object Detection for Autonomous Drone Control

Abstract
Unmanned Aerial Vehicles (UAVs) are more crucial today than ever before, finding application across various domains, with agriculture being a particularly important sector they serve. Their potential is far from exhausted, which makes research in this field essential. Applications such as wildlife monitoring and assessing the condition of plants and soils are already in practical use. Emerging tasks in agriculture are being explored through ongoing research projects, like the classification of strawberry maturity, which contributes to the ease of harvesting and predictability of profits. Central to all these applications is object detection, facilitating rapid decision-making. To enable this, real-time processing of image data is needed. To guarantee real-time capabilities, hardware-accelerated processing is employed. This involves leveraging software tools developed by Xilinx for AI workloads like object detection, working in conjunction with DPUs on FPGAs. Additionally, this work embraces a prototyping approach, offering cost-effectiveness and risk mitigation. The Robomaster Tello Talent drone serves as the prototype platform. The object detection implementation developed on this drone was successfully transferred to a larger application drone. With further developed object detection models, this system architecture can be easily extended to other use cases in the future.