Accéder directement au contenu Accéder directement à la navigation
Nouvelle interface
Communication dans un congrès

Performance of Recent tiny/small YOLO Versions in the Context of Top-view Fisheye Images

Benoit Faure 1 Nathan Odic 1 Olfa Haggui 2 Baptiste Magnier 1 
2 IDH - Interactive Digital Humans
LIRMM - Laboratoire d'Informatique de Robotique et de Microélectronique de Montpellier
Abstract : With the spreading of the computer vision field, human de- tection and tracking are problems more relevant than ever. However, due to the complexity of fisheye images, current lightweight detection models show difficulty when processing them. The aim of this article is to com- pare the performance on fisheye images of current real time detection solutions, specifically YOLOv3-tiny, YOLOv4-tiny and YOLOv5-small. Experiments carried out using a top-view fisheye camera, show faster performance but the very poor detection quality from YOLOv4-tiny. YOLOv5-small, while being slightly slower, gives a far better detection than both other solutions. The database created for this paper is avail- able online. In conclusion, the current review shows YOLOv5-small is the best solution out of the 3 reviewed in a fast, real time, fisheye application.
Type de document :
Communication dans un congrès
Liste complète des métadonnées

https://hal.mines-ales.fr/hal-03676679
Contributeur : Administrateur IMT - Mines Alès Connectez-vous pour contacter le contributeur
Soumis le : mardi 24 mai 2022 - 11:04:40
Dernière modification le : vendredi 4 novembre 2022 - 08:12:23
Archivage à long terme le : : mardi 30 août 2022 - 10:24:54

Fichier

performance_YOLO_fisheye_HAL.p...
Fichiers produits par l'(les) auteur(s)

Identifiants

  • HAL Id : hal-03676679, version 1

Citation

Benoit Faure, Nathan Odic, Olfa Haggui, Baptiste Magnier. Performance of Recent tiny/small YOLO Versions in the Context of Top-view Fisheye Images. ISHAPE 2022 - 1st International Workshop on Intelligent Systems in Human and Artificial Perception, May 2022, Lecce, Italy. ⟨hal-03676679⟩

Partager

Métriques

Consultations de la notice

45

Téléchargements de fichiers

146