A Linear Method for Virtual Object Synthesis and Camera-Pose Estimation from Three Affine Views

Abstract

We propose a linear algorithm that is useful for realizing the geometric registration between the views of real scene and virtual object in the image-based rendering framework. It can handle, in a unified framework, the novel view synthesis of virtual object based on the 3 view matching constraints and the recovery of camera pose that is necessary for the base image selection. Feasibility of the algorithm is demonstrated by using ground-truthed synthesized data and real scene data.