Détection temporelle de saillance dynamique dans des vidéos par apprentissage profond

Abstract : We address the problem of motion saliency in videos. More precisely, we aim to determine at each time step if an image can be classified as salient or not according to its motion content. An image will be detected as salient if it contains objects whose motion departs from its spatio-temporal context. The proposed approach handles situations with a mobile camera and involves a deep learning stage. Several variants are proposed and compared. Temporal saliency detection is relevant for applications that require to trigger alerts or to monitor dynamic behaviours from videos. Experiments on real videos demonstrate that the proposed methods can provide accurate classification.
Document type :
Conference papers
Complete list of metadatas

Cited literature [35 references]  Display  Hide  Download

https://hal.inria.fr/hal-01926351
Contributor : Léo Maczyta <>
Submitted on : Monday, November 19, 2018 - 10:51:21 AM
Last modification on : Wednesday, November 28, 2018 - 1:14:13 AM
Long-term archiving on: Wednesday, February 20, 2019 - 1:18:12 PM

File

Détection_Saillance.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : hal-01926351, version 1

Citation

Léo Maczyta, Patrick Bouthemy, Olivier Le Meur. Détection temporelle de saillance dynamique dans des vidéos par apprentissage profond. RFIAP 2018 - Reconnaissance des Formes, Image, Apprentissage et Perception, Jun 2018, Marne-la-Vallée, France. pp.1-8. ⟨hal-01926351⟩

Share

Metrics

Record views

180

Files downloads

127