Trajectory based Primitive Events for learning and recognizing Activity

Abstract : This paper proposes a framework to recognize and classify loosely constrained activities with minimal supervision. The framework use basic trajectory information as input and goes up to video interpretation. The work reduces the gap between low-level information and semantic interpretation, building an intermediate layer composed Primitive Events. The proposed representation for primitive events aims at capturing small meaningful motions over the scene with the advantage of been learnt in an unsupervised manner. We propose the modelling of an activity using Primitive Events as the main descriptors. The activity model is built in a semi-supervised way using only real tracking data. Finally we validate the descriptors by recognizing and labelling modelled activities in a home-care application dataset.
Document type :
Conference papers
Second IEEE International Workshop on Tracking Humans for the Evaluation of their Motion in Image Sequences (THEMIS2009), Sep 2009, Kyoto, Japan. 2009
Liste complète des métadonnées


https://hal.inria.fr/inria-00503209
Contributor : Guido Pusiol <>
Submitted on : Friday, July 16, 2010 - 7:48:58 PM
Last modification on : Friday, July 16, 2010 - 9:12:31 PM
Document(s) archivé(s) le : Friday, October 22, 2010 - 3:24:24 PM

File

PID967691.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : inria-00503209, version 1

Collections

Citation

Guido Pusiol, François Bremond, Monique Thonnat. Trajectory based Primitive Events for learning and recognizing Activity. Second IEEE International Workshop on Tracking Humans for the Evaluation of their Motion in Image Sequences (THEMIS2009), Sep 2009, Kyoto, Japan. 2009. <inria-00503209>

Share

Metrics

Record views

287

Document downloads

144