Skip to Main content Skip to Navigation
Preprints, Working Papers, ...

Représentations hiérarchiques et discriminantes pour la reconnaissance des formes, l'identification des personnes et l'analyse des mouvements dans les séquences d'images

Nicolas Thome 1
1 imagine - Extraction de Caractéristiques et Identification
LIRIS - Laboratoire d'InfoRmatique en Image et Systèmes d'information
Abstract : The field of human motion analysis by means of computer vision technologies has known a huge increase over the last decade. We propose here to tackle the problem in the following aspects : Licence Plate Recognition (LPR) and Human Motion Analysis. The proposed approach for LPR decomposes into the following steps: plate detection, binarization, character segmentation, shape classification and temporal averaging. Concerning Human Motion Analysis, our approach can be considered intermediate with respect to the Top Down and Bottom Up classes, When the tracking can be successfully achieved by a simple region association strategy, we analyse the shape silhouette at a finer scale, aiming at detecting and labelling body parts. The body parts labelling makes it possible to update an appearance model for each limb, capturing shape, color and texture properties. The appearance model is used in difficult situations to identify people. Concerning human motion analysis, we propose an approach dedicated to fall detection, whose contribution decomposes into two points. Firstly, we propose to analyse the relevance of a proposed verticality detector, computed from an image feature and that devoted to discriminating standing from lying poses. The second contribution corresponds to the application of a Hierarchical Hidden Markov Model (HHMM) to classify the pose sequences. The different proposed approaches for License Plate recognition or human motion analysis have proved the efficiency of Bottom Up approaches for a real time purpose, and pointed out the importance of generatives back propagations to revolve ambiguous situations.
Document type :
Preprints, Working Papers, ...
Complete list of metadatas

https://hal.archives-ouvertes.fr/hal-01463008
Contributor : Équipe Gestionnaire Des Publications Si Liris <>
Submitted on : Thursday, February 9, 2017 - 12:22:36 PM
Last modification on : Wednesday, October 14, 2020 - 4:15:54 AM

Identifiers

  • HAL Id : hal-01463008, version 1

Citation

Nicolas Thome. Représentations hiérarchiques et discriminantes pour la reconnaissance des formes, l'identification des personnes et l'analyse des mouvements dans les séquences d'images. 2007. ⟨hal-01463008⟩

Share

Metrics

Record views

211