REAL-TIME UNMANNED AERIAL VEHICLE DETECTION SYSTEM USING YOLOV8 AND SSDLITE ON NVIDIA JETSON NANO
Muhammad Tanta Rivansyah, Dr. Raden Sumiharto, S.Si., M.Kom
2025 | Skripsi | ELEKTRONIKA DAN INSTRUMENTASI
The increasing use of unmanned aerial vehicles (UAVs) across multiple industries has raised critical concerns regarding safety and security, highlighting the need for reliable detection systems. Traditional methods, often fall short, especially on devices with constrained computational resources. To address this issue, a computer vision approach using an edge device is proposed. This study presents a real-time, vision-based UAV detection system implemented on the NVIDIA Jetson Nano.
This system leverages deep learning algorithms, specifically YOLOv8 and SSDLite, to identify UAVs through bounding box outputs. YOLOv8n and SSDLite MobileNetV2 are selected for their efficiency and accuracy, making them suitable for deployment on a low-power device. The training and validation processes utilize a dataset sourced from a prior study, comprising 51,446 images for training and 2625 images for validation. The final models are optimized and deployed on the Jetson Nano, taking advantage of TensorRT to improve inference speed.
Experimental results show that the YOLOv8n model reaches a mean Average Precision (mAP) of 85%, while SSDLite MobileNetV2 achieves 69%. When tested on the Jetson Nano, YOLOv8n performs real-time detection at 19 FPS with 34 ms latency, and SSDLite runs at 14 FPS with 43 ms latency, demonstrating the feasibility of deploying lightweight AI models for UAV detection on embedded systems.
Kata Kunci : UAV Detection, YOLOv8, SSDLite, NVIDIA Jetson Nano