Avoiding Degeneracy for Line-based Visual SLAM Using Vanishing Point Re-projection

Abstract

We solve the degeneracy problem of line-based visual Simultaneous Localization and Mapping (SLAM) using vanishing point re-projection. Line features are used to supplement point features to cope with adverse environment. And the line features also enables environment recognition through structural map information. However, due to degeneracy, inaccurate mapping occurs. This is because the direction vector of the line cannot be accurately estimated during the triangulation process. In addition, the line re-projection error of existing algorithms does not correct the direction vector. To solve this problem, we propose a new constraint that observes the vanishing point in 2D image and compares it with the calculated vanishing point estimation of the 3D line. Experimenting with this on a public dataset shows that the mapping result is improved.