High Performance Non-linear Motion Blur

Abstract

Motion blur is becoming more common in interactive applications such as games and previsualization tools. Here, a common strategy is to approximate motion blur with an image-space post-process, and many recent approaches demonstrate very efficient and high-quality results [Sou13,GMN14]. Unfortunately, all such approaches assume underlying linear motion, and so they cannot approximate non-linear motion blur effects without significant visual artifacts.We present a new motion blur post-process that correctly treats the case of non-linear motion (in addition to linear motion) using an efficient curve-sampling scatter approach. We simulate plausible non-linear motion blur in 4ms at 1920 1080 and our approach has many desirable properties: its cost is independent of geometric complexity, it robustly estimates blurring extents to avoid typical over- and under-blurring artifacts, it supports unlimited motion magnitudes, and it is less noisy than existing techniques.