Spherical PTAM : a versatile SLAM for spherical video

The visual Simultaneous Localization and Mapping (SLAM) working with spherical video gives many advantages of a wide field of view. However, the conventional visual SLAM approaches are not directly applicable to the spherical videos due to the distortion introduced by equirectangular projection. Feature points extracted from spherical video frames are pervaded by spherical distortion. The spherical distortion should be properly handled to apply classical planar SLAM algorithm. For the versatile and accurate map construction, a PTAM for the spherical video requires an effective means to exclude the spherical distortion. This article proposes a novel visual SLAM method for 360-degree spherical video based on the monocular Parallel Tracking and Mapping (PTAM) approach. The propose method performs tracking and mapping with high accuracy through the idea of calculation and matching undistorted patches on the dual projection plane model. The experimental results show the proposed method gives high accuracy and stable performance.