ImPL-VIO: An Improved Monocular Visual-Inertial Odometry Using Point and Line Features

Most of the visual-inertial navigation systems (VINS) that use only point features usually work well in regular environment, but decay in low-texture scenes. Meanwhile, those systems rarely construct environmental map with structural information. In this paper, an improved tightly-coupled monocular visual-inertial odometry (ImPL-VIO) is developed. The whole system is composed of point and line feature tracking, inertial measurements processing, pose estimator and loop closure detection. For the better use of monocular line observations in the sliding window based pose estimator, an improved line triangulation algorithm is proposed after a detailed analysis of error sources. In addition, we, for the first time, employ the closest point (CP) representation for spatial lines to optimization-based VINS system, and derive the corresponding Jacobians analytically. Finally, simulation and real-world experiments are conducted to validate the proposed system.