Skip to Main content Skip to Navigation
Theses

Méthodes par ensembles de niveaux et modes conditionnels itérés pour la segmentation vidéo

Abstract : This thesis is devoted to the study of a computer vision problem and of two video surveillance problems. We propose a method for moving objects detection in a video sequence; it is based on the apparent velocity predetermination and on a shape optimisation problem. For other moving objects detection and tracking models, we propose to use the iterated conditional modes discrete algorithm which is renowned to be very fast and allows to decrease the computing time of the continuous algorithms when it is combined to them. For video surveillance, we seek to estimate crowd density and to detect abnormal behaviour in parisian subway environment. We propose a crowd density estimation based on the computation of curvature on the image. Abnormal behaviour detection is performed by a research of the modes in the histogram of the apparent velocity directions.
Complete list of metadatas

Cited literature [2 references]  Display  Hide  Download

https://tel.archives-ouvertes.fr/tel-00008038
Contributor : Florent Ranchin <>
Submitted on : Friday, March 4, 2005 - 11:10:35 AM
Last modification on : Wednesday, September 23, 2020 - 4:30:46 AM
Long-term archiving on: : Monday, September 20, 2010 - 12:20:06 PM

Identifiers

  • HAL Id : tel-00008038, version 2

Collections

Citation

Florent Ranchin. Méthodes par ensembles de niveaux et modes conditionnels itérés pour la segmentation vidéo. Interface homme-machine [cs.HC]. Université Paris Dauphine - Paris IX, 2004. Français. ⟨tel-00008038v2⟩

Share

Metrics

Record views

338

Files downloads

1567