Spherical Camera Localization by Color Difference Minimization Using 3D Model of the Environment

The goal of this study is to propose an occlusion-robust spherical camera localization method using 3D model of the environment. Instead of using specific features, which may cause low estimation accuracy in environment with occlusion, the color information of all pixels in images is used for the estimation. Camera pose is estimated by minimizing the color difference of the taken image and the image generated using former frame and 3D model. To reduce the influence caused by occlusion, a robust evaluation function is used in the estimation process. The effectiveness of the proposed method was confirmed by experimental tests.