Agents in Social Science
115
to when they are part of a crowd. Sociological and behavioral simulation of
people in closed environments allows studying relationships between different
people from a social perspective. This can often display the existence of hierarchy inside a group, such as leadership and member relations among the
individuals. Most crowd models have investigated behavior of one pedestrian
within a crowd. However, group pattern formation and other types of social
relationships can be extensively studied in crowds using agent-based models.
Early examples can be seen by Reynolds [160], where he presented distributed
behavioral model to produce flocking behavior.
Person agents in crowds can be goal-directed, reactive or opportunistic.
Modelers have to program this preference into agents. Modeling crowd behavior requires a large amount of data analysis with different densities, numbers and heterogeneous behaviors. Most crowd behavior analysis is done using
video tracking software and hindered by additional entities in scenes such as
loose clothing, carrying umbrellas, bags or packages.
In FLAME, a crowd model was programmed as follows:
1. Initialize agents in scene (such as at entrance of a corridor).
• Initialize 50 agents.
• Randomly add people in a group to generate various group sizes.
• Randomly generate families and assign ages between 4 to 100.
• In groups, assign a leader.
• Assign a destination exit for each agent group.
2. Post agent location for other agents to read.
3. Choose one of the following behaviors at random:
• If agent is in group, collision avoidance to mediate movement.
• If the agent is out of the group’s circle, bring agent back towards
group.
• If in a group, the agent follows the leader.
• The agent goes to nearest shop.
4. Walk all agents towards the exit.
The pedestrians are simulated as individual person agents walking in the
crowd.
Person Agent. Contains id, x, y, speed, gender, exit no, is a leader or not,
which group do I belong to and other variables such as weights associated
to prevent collisions and walk towards goals.
Generator Agent. Contains the total number of persons generated. This
agent is necessary to create agents at corridor entry points in the simulation.
115
to when they are part of a crowd. Sociological and behavioral simulation of
people in closed environments allows studying relationships between different
people from a social perspective. This can often display the existence of hierarchy inside a group, such as leadership and member relations among the
individuals. Most crowd models have investigated behavior of one pedestrian
within a crowd. However, group pattern formation and other types of social
relationships can be extensively studied in crowds using agent-based models.
Early examples can be seen by Reynolds [160], where he presented distributed
behavioral model to produce flocking behavior.
Person agents in crowds can be goal-directed, reactive or opportunistic.
Modelers have to program this preference into agents. Modeling crowd behavior requires a large amount of data analysis with different densities, numbers and heterogeneous behaviors. Most crowd behavior analysis is done using
video tracking software and hindered by additional entities in scenes such as
loose clothing, carrying umbrellas, bags or packages.
In FLAME, a crowd model was programmed as follows:
1. Initialize agents in scene (such as at entrance of a corridor).
• Initialize 50 agents.
• Randomly add people in a group to generate various group sizes.
• Randomly generate families and assign ages between 4 to 100.
• In groups, assign a leader.
• Assign a destination exit for each agent group.
2. Post agent location for other agents to read.
3. Choose one of the following behaviors at random:
• If agent is in group, collision avoidance to mediate movement.
• If the agent is out of the group’s circle, bring agent back towards
group.
• If in a group, the agent follows the leader.
• The agent goes to nearest shop.
4. Walk all agents towards the exit.
The pedestrians are simulated as individual person agents walking in the
crowd.
Person Agent. Contains id, x, y, speed, gender, exit no, is a leader or not,
which group do I belong to and other variables such as weights associated
to prevent collisions and walk towards goals.
Generator Agent. Contains the total number of persons generated. This
agent is necessary to create agents at corridor entry points in the simulation.