Paper Key : IRJ************475
Author: Atharv Khonde,Piyush Bagde,Kaiwalyani Gorde,Prof. Anirudh Bhagwat
Date Published: 05 Nov 2024
Abstract
Abstract : For search and rescue (SAR) operations, person detection in aerial images is essential, as it impacts the speed and effectiveness of finding people in trouble. This review studies the use of YOLOv8 nano, a deep learning model optimised for resource efficiency and real-time processing, to person detection in SAR, with an emphasis on its implementation. Conventional human being methods of detection suffer in the diverse and uncertain circumstances that most SAR missions encounter. Deep learning techniquesparticularly CNNshave demonstrated significant improvements in accuracy and versatility. For SAR operations where time is of the essence, the most recent version, YOLOv8, provides increased speed and efficiency. Its nano configuration, optimized for resource-constrained devices, makes it especially appropriate for nano drones, which are essential for accessing difficult terrain and disaster zones where larger UAVs may be not feasible. The paper highlights the advantages and disadvantages of different models by analysing previous research on deep learning for SAR person detection. Although research utilising YOLOv5 and SlimYOLOv3 shows this model family's potential for aerial picture analysis and quick detection times, issues with accuracy in a variety of situations and the requirement for strong training datasets still need to be addressed. For example, while previous studies utilising Faster R-CNN showed promise, it struggled to recognise small objects, which is a vital requirement in SAR.
DOI Requested