B. White and L. Bölöni

Automatic Analysis of Embodied Team Actions


Cite as:

B. White and L. Bölöni. Automatic Analysis of Embodied Team Actions. In Workshop on Plan, Activity, and Intent Recognition (PAIR-2009) at IJCAI-2009, 2009.

Download:

Download 

Abstract:

We describe a system which, building on previous team action recognition systems, performs a more in-depth analysis of an ongoing team action executed by a group of embodied agents. The system relies on team action states with human understandable semantics, estimates the current state and is able to make predictions or identify fringe cases such as incomplete or incorrectly executed team actions. The representation of the team action relies on a dynamic Bayesian network (DBN). We perform reasoning over the DBN using a sampling-importance-resampling particle filter. As a methodological illustration, we describe the process of model building for the bounding overwatch team action. We experimentally test our approach using data acquired from video recordings, and measure the systems' ability to recognize a team action and to estimate the current state.

BibTeX:

@inproceedings{White-2009-PAIR,
  author = "B. White and L. B{\"o}l{\"o}ni",
  title = "Automatic Analysis of Embodied Team Actions",
  booktitle = "Workshop on Plan, Activity, and Intent Recognition (PAIR-2009) at IJCAI-2009",
  year = "2009",
  abstract = {
    We describe a system which, building on previous team action
    recognition systems, performs a more in-depth analysis of an ongoing
    team action executed by a group of embodied agents. The system
    relies on team action states with human understandable semantics,
    estimates the current state and is able to make predictions or
    identify fringe cases such as incomplete or incorrectly executed
    team actions. The representation of the team action relies on a
    dynamic Bayesian network (DBN). We perform reasoning over the DBN
    using a sampling-importance-resampling particle filter. As a
    methodological illustration, we describe the process of model
    building for the bounding overwatch team action. We experimentally
    test our approach using data acquired from video recordings, and
    measure the systems' ability to recognize a team action and to
    estimate the current state.
  }
}

Generated by bib2html.pl (written by Patrick Riley, Lotzi Boloni ) on Fri Oct 06, 2017 18:15:24