GPU-assisted Multi-field Video Volume Visualization

Abstract

GPU-assisted multi-field rendering provides a means of generating effective video volume visualization that can convey both the objects in a spatiotemporal domain as well as the motion status of these objects. In this paper, we present a technical framework that enables combined volume and flow visualization of a video to be synthesized using GPU-based techniques. A bricking-based volume rendering method is deployed for handling large video datasets in a scalable manner, which is particularly useful for synthesizing a dynamic visualization of a video stream. We have implemented a number of image processing filters, and in particular, we employ an optical flow filter for estimating motion flows in a video. We have devised mechanisms for combining volume objects in a scalar field with glyph and streamline geometry from an optical flow. We demonstrate the effectiveness of our approach with example visualizations constructed from two benchmarking problems in computer vision.