comp870 - Project


For my project, I will implement fast camera motion blurring from sample reprojection. Reprojecting a sample is fast compared to tracing a new sample, so reprojection is used to reduce intersection tests needed when tracing a scene, thus increasing update rates. Since reprojection is a relatively cheap operation, I will use reprojected samples to estimate the motion blur from camera motion. This will be similar to the use of reprojected samples in frameless rendering, but instead of using reprojections to decrease intersection tests, they will be used to create a higher quality image.

Since computer generated images are discrete and often independent, they lack natural motion blur between frames. Without blur, the frames must be updates at high rates to give the illusion of motion. By using motion blur between frames, it is possible that the smooth motion could be perceived with lower updates rates.

As the camera is in motion, the scene will be rendered. While the next frame is rendering, the next camera transform will be used to reproject samples. Samples will be reprojected in the image space and interpolated based on the degree of motion in image space. This will result in a motion blurred image with low cost. This blurred image will be paired with the next frame to produce the final output.