Accelerated Projector Compensation Framework using Differentiable Rendering

Abstract

A general scenario of a projection mapping is a projector projects the input image on the surface, and then camera captures the projected image on the surface. When the surface is ideal(white and planar), a camera captures the image projected onto surface without any visual distortion. However, in practice, the surface can be non-planar and colored, which cause unexpected visual distortion. It restricts the viewing experience to viewer. Projector compensation is the optimization problem that minimizes the difference between the camera captured image and the target image by adjusting the projector input image. The Differentiable rendering base method shows the best result among proposed methods. Differentiable rendering base method model the real projection mapping with rendering function and optimizing the projector input image based on differentiable rendering. We propose and demonstrate an improved framework that accelerates the speed of image generation time without a significant decrease in quality, which is the limitation of the previous differentiable rendering base method.