Tracking on full-view image for camera motion estimation based on spherical model

In this paper we propose a method of tracking the surrounding markers in a spherical image for camera motion estimation of VR (Virtual Reality) applications. The core algorithm is a modification of some known planar image binary feature descriptors. A pattern is defined first. Then, using the orientation estimation, the rotated pattern for each feature point is determined. However, the proposed method is based on spherical model, in which the point coordinate is presented as 3D unit coordinate and the distance between two neighbor points is not the same because of the icosahedron subdivision strategy. To solve this problem, we resort to spherical polar coordinate and spherical rotation. Finally, by solving the PNP problem, the spherical camera motion is estimated. The experiments show our method for tracking on spherical images is effective and efficient.