Aerial object detection analysis: Challenges and preliminary results

Informatics Department, Sanata Dharma University, Yogyakarta, Indonesia

^* Corresponding author: a.m.polina@usd.ac.id

Abstract

Computer vision allows computers to retrieve information from images, videos, and other visual inputs. Unmanned Aerial Vehicle (UAV) technology is also used to assist computer vision in collecting image data from the air. This paper aims to perform tree object detection using UAVs by capturing images perpendicularly from above the object. Image data was collected from around Sleman Yogyakarta using DJI Pro 3 from 5 to 12 July 2023. A total of 162 images were used as a dataset. The YOLOv8n model was implemented to 162 images as the training and validation data. Next, 12 other images were used as testing data. The results showed that YOLOv8n could detect trees well from above. The confidence value of the testing dataset with the appropriate image capture is more than 80%. As a deep learning algorithm for object detection, the YOLO model can perform object detection quickly and accurately. The subsequent research will focus on analyzing the implementation of object detection using the YOLO algorithm to measure open green areas.