Modeling and Synthesis of Crowd Movement Based on a Low-dimensional Manifold Representation

Project: Research

View graph of relations

Researcher(s)

Description

Due to the increasing need to perform crowd scene generation in application areas such as the synthesis of sports events audiences and street crowds in movie production, and battalion scene modelling in game design, effective and efficient approaches for constructing an accurate model for representing group movements become more and more important. In view of the large number of degrees of freedom of the time varying motion field associated with crowd movement, the researchers propose to search for an optimal low-dimensional manifold to characterize the crowd motion pattern. Specifically, each snapshot of this field can be represented as a point on the manifold, such that the complete time-varying field can be modelled as a trajectory on the surface. Sampling can then be effectively performed on this low-dimensional manifold to obtain a new spatiotemporal trajectory to create a novel yet realistic crowd motion field, which is then used to drive the motion of a set of virtual characters. The proposed approach thus alleviates the problem of artificiality by adopting an example-based learning approach to determine the motion model, instead of explicitly specifying the model properties, and the problem of scalability by adopting a global crowd motion field instead of associating each virtual character with its own motion model.

Detail(s)

Project number9041263
Grant typeGRF
StatusFinished
Effective start/end date1/01/081/03/11