Model of attention for a virtual agent able to interact with a user through body movements - DUMAS - Dépôt Universitaire de Mémoires Après Soutenance Accéder directement au contenu
Mémoires Année : 2014

Model of attention for a virtual agent able to interact with a user through body movements

Résumé

This report presents the work during my internship concerning the creation of a model of attention for a virtual agent capable of interacting with a human through body movements. Attention is defined as the process of filtering on the intelligent entities senses. Depending on the stimuli the attention is characterized as bottom up (external stimuli) or top down. Different models of attention argue on the time a perception filter acts on attended or unattended information or introduce the notion of attentional capacity and argue that the filtering relates to the semantic meaning of the perceived information. Also, specific behaviour expressed by body movements, gestures and gaze is related to attention that is an intrinsic human mechanism. To create believable virtual agents that resemble human, implementation of attention mechanisms is needed. The existing computational models are grouped into methods that perform either visual search to find points of interest or semantic processing or implement overt behaviour or combine methods that are used for expressive embodied conversational agents. Based on the previous work, we propose some ideas about the conception of our model. We think that a combination of bottom up and top down method along with a vision field filtering for the agent that depend on its head orientation in order to show a more believable behaviour. A module of Attention was created as a part of a multi-module architecture virtual agent architecture. The modules communicate through messages. The Attention module consists of a vision field filter, a motion calculation, a state machine, a strategy pattern. The vision field filters whether the user is visible or not to the agent. Then the motion calculation takes into account the visible parts of the user that move in order to calculate a level of attention for the agent. Then depending on the input from the vision filter and the motion calculation a state is calculated and an action (message is sent) is taken.

Mots clés

Fichier principal
Vignette du fichier
mri20132014_submission_80.pdf (4.04 Mo) Télécharger le fichier
Loading...

Dates et versions

dumas-01088827 , version 1 (28-11-2014)

Licence

Paternité

Identifiants

  • HAL Id : dumas-01088827 , version 1

Citer

Panagiotis Mavridis. Model of attention for a virtual agent able to interact with a user through body movements. Graphics [cs.GR]. 2014. ⟨dumas-01088827⟩
304 Consultations
267 Téléchargements

Partager

Gmail Facebook X LinkedIn More