Detecting Swimmers in Unconstrained Videos with Few Training Data - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Machine Learning and Data Mining for Sports Analytics (MLSA) Année : 2021

Detecting Swimmers in Unconstrained Videos with Few Training Data

Résumé

In this work, we propose a method to detect swimmers in unconstrained swimming video, using a Unet-based model trained on a small dataset. Our main motivation is to make the method accessible without spending much time or money in annotation or computation while maintaining performance. As a result, our model reaches topperformances in detection without requiring for intrusive sensors. The swimming videos can be recorded from various locations with different settings (distance and angle to the pool, light conditions, reflections, camera resolution), which alleviates a lot of the usual video capture constraints. Every algorithm described in the paper is accessible online at https://github.com/njacquelin/swimmers detection.
Fichier principal
Vignette du fichier
Detecting Swimmers in Unconstrained Videos with Few Training Data.pdf (5.99 Mo) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-03358375 , version 1 (29-09-2021)

Identifiants

  • HAL Id : hal-03358375 , version 1

Citer

Nicolas Jacquelin, Romain Vuillemot, Stefan Duffner. Detecting Swimmers in Unconstrained Videos with Few Training Data. Machine Learning and Data Mining for Sports Analytics, Sep 2021, Ghand, Belgium. ⟨hal-03358375⟩
183 Consultations
314 Téléchargements

Partager

Gmail Facebook X LinkedIn More