Discrete relative states to learn and recognize goals-based behaviors of groups

  • Authors:

  • Jérémy Patrix;Abdel-Illah Mouaddib;Simon Le Gloannec;Dafni Stampouli;Marc Contat

  • Affiliations:

  • GREYC (UMPR 6072), Caen, France;GREYC (UMPR 6072), Caen, France;CASSIDIAN, Val-de-Reuil, France;CASSIDIAN, Val-de-Reuil, France;CASSIDIAN, Val-de-Reuil, France

  • Venue:
  • Proceedings of the 2013 international conference on Autonomous agents and multi-agent systems
  • Year:
  • 2013

Quantified Score

Hi-index 0.00

Visualization

Abstract

In a crisis management context, situation awareness is challenging due to the complexity of the environment and the limited resources available to the security forces. The different emerging threats are difficult to identify and the behavior of the crowd (separated in groups) is difficult to interpret and manage. In order to solve this problem, the authors propose a method to detect threat and understand the situation by analyzing the collective behavior of groups inside the crowd and detecting their goals. This is done according to a set of learned, goal-based, group behavior models and observation sequences of the group. The proposed method computes the group estimated state before using Hidden Markov Model to recognize the goal by the group behavior. A realistic emergency scenario is simulated to demonstrate the performance of the algorithms, where a suicide-bomber wearing a concealed bomb enters a busy urban street. The proposed algorithms achieve the detection of the dangerous person in the crowd, in order to raise an alert and also predict casualties by identifying which groups did not notice the threat. Complex Event Processing is used to compare and evaluate the results. The algorithms were found more precise and more tolerant to noisy observations.