Real-Time Human Detection in Marine Environment Using Deep Learning on Edge Devices - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2022

Real-Time Human Detection in Marine Environment Using Deep Learning on Edge Devices

Résumé

Artificial intelligence (AI) techniques based on deep learning provide robust solutions to detect and locate objects. The achieved performance prove the relevance of convolution neural networks (CNNs) in circumventing existing computer vision challenges. The goal of this work is to exploit the advantages of AI-based methods in detection of floating humans in open water to aid marine search and rescue missions. This has the potential to save lives while simultaneously saving efforts and expenses. In this work, we explore the use of You Only Look Once (YOLO) in detecting humans in maritime environment. A custom dataset is used to train the available YOLOv4 models. The trained models are assessed using recognized evaluation metrics. In addition, the inference speed is targeted towards embedded low-power hardware edge devices. The obtained results reveal that YOLOv4 can identify persons in a marine environment in real time with acceptable accuracy and precision.
Fichier principal
Vignette du fichier
GDRSoC2022.pdf (112.28 Ko) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-03698760 , version 1 (19-06-2022)

Identifiants

  • HAL Id : hal-03698760 , version 1

Citer

Mostafa Rizk, Amer Baghdadi, J-Ph Diguet. Real-Time Human Detection in Marine Environment Using Deep Learning on Edge Devices. GDR SoC2: Groupe de recherche System on Chip – Systèmes embarqués et Objets Connectés, Colloque National, Jun 2022, Strasbourg, France. ⟨hal-03698760⟩
67 Consultations
108 Téléchargements

Partager

Gmail Facebook X LinkedIn More