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

Goats monitoring at the pasture scale combining neural network and time-lapse cameras

Abstract : This paper presents a system for animal monitoring on pasture that combines a state of the art object detection algorithm named YOLO, and low cost time-lapse cameras. The cameras network is taking pictures of the flock every 30 seconds and several algorithms are used to automatically detect the animals and estimate their positions on the pasture. The system was tested on a 1300m2 pasture with 18 creole goats and 22 kids. Based on 1000 pictures selected randomly, the animal detection sensitivity and precision was estimated to 81.3% and 97.1%. Unlike GPS, where one sensor per animal is needed, the system can be used to provide quantitative information on the entire flock using only three passive sensors. We used the system to study the flock during two grazing weeks on the same pasture, with one month interval. We showed that from one week to another, the flock spatial distribution was highly correlated. The correlation is not constant but decreases gradually during the week. More generally, the flock positions can be used for quantitative studies on animal behavior, animal health and welfare, or to improve pasture management.
Liste complète des métadonnées

https://hal.inrae.fr/hal-02737985
Déposant : Migration Prodinra <>
Soumis le : mardi 2 juin 2020 - 20:07:58
Dernière modification le : vendredi 12 juin 2020 - 10:43:26

Identifiants

  • HAL Id : hal-02737985, version 1
  • PRODINRA : 479486

Collections

Citation

Mathieu Bonneau, Jehan-Antoine Vayssade, Willy Troupé, Rémy Arquet. Goats monitoring at the pasture scale combining neural network and time-lapse cameras. European Conference on Precisions Livestock Farming (ECPLF), Aug 2019, Cork, Ireland. ⟨hal-02737985⟩

Partager

Métriques

Consultations de la notice

8