In our work, we address the problem of simulating crowds. Modeling crowd behavior in a realistic manner is of great significance in the motion picture industry. Current approaches, employed by industry heavyweights such as ILM and made use of in the making of the Lord of the Rings trilogy, are mainly based on hand-picked rules dependent on the scene being modeled. What one would ideally like to have is a generic framework which generates scenes given some input parameters which characterize the crowd behavior.
There has been some recent work on this problem of generating virtual crowds. Some of them tackle the problem using a leader-follower model [1], while others using flocking models [2] similar to those developed for modeling birds. The work which most resembles the problem definition we are addressing is that of Metoyer and Hodgins [3]. In their paper, they allow a user to first choose the spots at which obstacles lie and from where people start off from, and then plot out the approximate paths of each person in the crowd. The path taken by each person is then determined using learning based on a naive Bayes classifier.
The new features that we handle in our work in comparison with that of [3] are the following:
The specific scene that we concentrate on in our work is that of people walking along a pavement, with additional people entering the scene from a subway. We restrict our work to showing a top view of the scene, with people in the scene represented as circles, and do not address the issue of rendering the moving people.