Synthesizing Relative Radiance for Realistic Rendering of Virtual Objects in 3D Photo Collections

Abstract

We present a novel approach for accurate reconstruction of relative radiance of every surface in an augmented image-based 3D scene. By using unordered images with varying exposure of 3D Photo Collections and a 3D mesh of the scene geometry, the relative radiance of all surfaces in the scene is reconstructed with High Dynamic Range. This is performed by subdividing the surfaces of the scene into patches and calculating the HDR image estimate for each patch. The result is used to embed virtual objects into 3D Photo Collections with authentic lighting.