Implementation of YOLOV5 for Real-time Detection of Ships and Buoys using Jetson Nano and Deepstream on Autonomous Leisure Vessel

Prayoga, Ardi
Alfitho Surya Prabaswara
Adri Firmansya Sofyan
Susanto, Erwin
Muhammad Hablul Barri

Serambi Engineering
Vol. 9 No. 4 (2024): Oktober 2024

System detection on autonomous vehicle have become an increasingly important subject of research in recent years due to their potential to enhance safety and operational efficiency in the maritime sector. This research focuses on the implementation of YOLOv5 for real-time object detection, specifically foreign vessels and buoys, using Jetson Nano and Deepstream. The developed system utilizes serial communication between Jetson Nano and Arduino Mega 2560 to receive and display object detection data. Testing results show that the object detection system achieves 96.66% accuracy during the day and 90% at night. During the YOLOv5 model training, precision of 91.96%, recall of 77.69%, mAP_0.5 of 81.55%, and mAP 0.5:0.95 of 66.23% were obtained. This implementation enables autonomous vessels to detect and avoid objects in real-time, thereby improving safety and operational efficiency at sea.