An automatic crowd generation system using image processing techniques

In advertisements, movies, and virtual environment modeling, synthetic crowd generation and processing techniques are needed. Currently, the most common crowd generation techniques are simply overlaying pre-captured crowd images onto the background image. In this paper we introduce an automatic method to generate realistic crowds. The system handles the data of a person's image and its attributes in a database to merge with background image. The method includes: (1) zooming an image in/out, depending on the distance from virtual viewport to the center of an image; (2) blurring an image boundary; (3) shadowing an image of person with a virtual light source; and (4) mirroring an image. We also propose various distribution functions to make a realistic crowd model and give implementation results. Improvements in the system for virtual reality applications on the WWW are being studied.