Authoring Reusable Crowds using Semantics

Abstract

In the field of video surveillance, a large problem is often the acquiring of video feeds of specific scenarios as test or training data. Even if privacy and safety concerns are not taken into account, it is often much cheaper to generate a virtual scenario rather than record one using real cameras, especially when the camera placement was not as perfect as desired when shooting the scenario. While recent advances in the field of crowd simulation have led to more and more realistic behavior, little attention has been given to the potential reuse of the crowds simulated in these scenarios. Not having to redefine a crowd for every scenario would make the authoring of simulations much easier, as the reuse of the environments already does. This thesis introduces a new crowd and agent model, that emphasize the reusability of the crowd. By semantically enriching the entities in the environments with information on what they can provide, the agent model is decoupled from specific environments, thereby allowing the agents to walk around in other environments. The crowd model is furthermore split into two components, one of which does not depend on any specific environment instance, while the other component remains inevitably dependent on such instances, otherwise the crowds would not be able to ‘know’ where to enter the environment. As a proof of concept, a prototype simulator for the crowd and agent models is also introduced, along with graphical editors for them. To assess both the models and the simulator, four test scenarios are created using the new crowd and agent models. Although the crowds can be inserted in any environment, they will not automatically work in all environments if the authoring process was not done with reusability in mind. Therefore several guidelines are constructed from the results of the scenarios, that will help with the authoring of both reusable crowds and reusable environments. With the crowd model introduced in this thesis, and the help of the accompanying authoring guidelines, virtual crowds may be reused more easily. By reusing the crowds as well as the environments, virtual scenarios can be developed more easily, thereby allowing more time to be spent on the tools that the scenarios were meant to test or train.