A proposPersonnesRecherchePublicationsEvenementsProfil
A propos







Integration of Uncertainty in the Analysis of Dyadic Human Activities

Maryam Ziaeefard and Robert Bergevin

Abstract - Action analysis from video data has been attracting more and more attention in computer vision over the past decade. The main focus has been classifying videos into one of k action classes from fully observed videos. However, all of the time, intelligent systems are enforced to make decisions under uncertainty and based on incomplete information. This need motivates us to introduce the problem of analyzing the uncertainty associated with dyadic human activities and move to a new level of generality in the action analysis problem. Analyzing the uncertainty, here, refers to categorizing the likelihood of activities in trimmed video clips called timeslices which are extracted from the full video. To this intent, we exploit the state-of-the-art methods to extract interest points in time-slices and represent them. We also present an accumulative uncertainty to depict the uncertainty associated with partially observed videos for the task of early activity recognition. The experiments demonstrate the effectiveness of our framework in analyzing dyadic activities under uncertainty and in evaluating the performance of early activity recognition methods.


    author    = { Maryam Ziaeefard and Robert Bergevin },
    title     = { Integration of Uncertainty in the Analysis of Dyadic Human Activities },
    booktitle = { 13th Canadian Conference on Computer and Robot Vision CRV 2016 },
    address   = { Victoria, BC },
    year      = { 2016 },
    month     = { June }

Dernière modification: Jun 14 2016 4:53PM par MAZIA2


©2002-. Laboratoire de Vision et Systèmes Numériques. Tous droits réservés