ARTIST-DIRECTED VOLUMETRIC DYNAMIC VIRTUAL CAMERAS

摘要

A volumetric dynamic virtual camera system employs a radial basis function (RBF) component that can utilize non-uniform training datasets that are blended to provide interpolated camera parameters during application runtime based on a player's position within virtual volumes in a 3D space. During application development, an artist or developer can interactively author cameras that are finely tuned to appear just right and which provide the training data for runtime. The RBF component blends the training data during runtime as the player's position within the volume changes to produce camera parameters that the camera system uses to capture scenes for rendering on a display device. The result is an overall camera system that lets authors very quickly develop film-quality cameras that appear and behave more like fully dynamic cameras having significant intelligence. The cameras are volumetric because they can exist in all the virtual spaces exposed by the application.

主权项

1. An apparatus for capturing virtual scenes generated by an application, the scenes occurring in a virtual three-dimensional (3D) application space, the system comprising:
one or more processors; and a computer-readable memory storing instructions which, when executed by the one or more processors, implement
a radial basis function (RBF) component arranged for receiving a current position of a player in the virtual 3D application space as an input and generating camera parameters as an output responsively to the current player position, the RBF component generating the camera parameters by interpolating training data, the training data including a plurality of data pairs, each data pair including a trained player position within a portion of the virtual 3D application space and an associated set of one or more trained camera parameters,a camera system arranged for receiving the camera parameters from the RBF component during application runtime and, responsively to the camera parameters, capturing a scene of a portion of the virtual 3D application space that includes the player in the current position.