A reconstruction filter for plausible motion blur

Abstract

Figure 1: A dragon tracked by a dolly camera over a burning city, with motion blur result reconstructed in 3 ms at 1280×720 on GeForce 480 from a conventionally-rasterized framebuffer. This image contains both character and camera motion comprising zoom, pan, translation, deformation, and rotation. The full-resolution auxiliary depth and velocity input buffers are inset at reduced size. We visualize velocity with a coarse vector field over the fine velocity buffer observed when implementing the algorithm, in which (r, g, b) = (dx/dt, dy/dt, 0)/k + 0.5. AbstractThis paper describes a novel filter for simulating motion blur phenomena in real time by applying ideas from offline stochastic reconstruction. The filter operates as a 2D post-process on a conventional framebuffer augmented with a screen-space velocity buffer. We demonstrate results on video game scenes rendered and reconstructed in real-time on NVIDIA GeForce 480 and Xbox 360 platforms, and show that the same filter can be applied to cinematic post-processing of offline-rendered images and real photographs. The technique is fast and robust enough that we deployed it in a production game engine used at Vicarious Visions.

Real-Time Motion Blur Using Triangular Motion Paths

“…Therefore, the visibility problem in the space-time dimensions is improperly handled. Later, McGuire et al [5] use the scatteras-gather based approach to generate blurred images. Scatter operations are operations that scatter a pixel's color to its neighbor, and gather operations are operations that gather color samples from neighboring pixels.…”

“…Therefore, instead of scattering a pixel's color to its neighbor, a scatter-as-gather based approach gathers samples from neighboring pixels based on the current pixel's velocity. McGuire et al [5] generate two intermediate buffers from a velocity buffer. For each pixel, the first intermediate buffer (TileMax) stores a velocity vector that has the largest magnitude in every k×k tile; and the second intermediate buffer (NeighborMax) stores a velocity vector that has the largest magnitude from the neighboring pixels.…”

“…Consequently, result images have artifacts as objects move at different speeds and directions. Guertin et al [6] improve an algorithm described by McGuire et al [5] and produce high-quality blurred images by sampling along a maximum velocity vector and a pixel's velocity vector. Yet this algorithm does not completely solve the problem.…”

“…Yet this algorithm does not completely solve the problem. We refer readers to McGuire et al [5] for some older post-processing algorithms which use extruded geometry to render motion blur but those algorithms have the same visibility problem of postprocessing approach.…”

Simulated motion blur does not improve player experience in racing game

“…Advanced features, such as higher-order motion blur ] and handling of disocclusions [McGuire et al 2012], enable accurate rendition in real time but at the cost of increased computational complexity. It is useful to understand whether these advanced features benefit players' experience of the game or if they go unnoticed.…”