Abstract
Geometric structures such as lines and planes are relevant in SLAM, as they improve the map interpretability and usability for downstream tasks. Planar landmarks add structural constraints to the map optimization, which could improve the accuracy of camera pose estimates. However, does the latter really happen in practice? In this paper, we thoroughly evaluate the effect of adding planar constraints in monocular SLAM, both in simulated and real scenes. Our experiments show that, in practical use cases, the benefit of adding such planar constraint shows benefits for scene estimation but limited effect in the camera pose estimation.