Modelling of Group Behaviours within Crowd Simulations : An Agent Based Approach

Abstract

Various social groups such as families, friends and co-workers are contained within real world crowds. According to Cooley's classi cation, these social groups are categorised as primary groups and secondary groups, of which decisions of a person are more signi cantly a ected by primary groups in comparison to secondary groups. Therefore, primary groups could prove to be a signi cant dependency to the behaviours of the crowds. Even with their strong relationships and similarity, persons in a group may have di erent personal goals. Despite the large amount of research done in this area, there is a lack of a proper method for simulating groups of characters with di erent goals. An agent based behavioural model is proposed by this research as a solution to this problem focusing on simulating the decision making of humans in crowd characters. Character decisions are approximated upon three parameters: utilities, attractions to others and distances to goals. An algorithm is proposed for implementing this decision making function of the behavioural model. The proposed solution is evaluated for its realism and e ciency using quantitative methods. Decisions of real and simulated characters are compared when measuring realism, through an experiment with human subjects. Update rate of agents were compared against number of groups, characters and goals as measures of e ciency of the solution. The outcomes of evaluations shown that the proposed model is capable of simulating characters of groups with a moderate level of realism. Moreover, it was found that number of groups and number of goals per character a ected the performance only slightly. It was concluded that the proposed behavioural model performed fairly well as a solution for simulating groups of characters who have di erent goals.