A Fast Monocular Visual–Inertial Odometry Using Point and Line Features

Most SLAM systems using point feature reduce the pose accuracy or crash due to the small number of point features are extracted under the light-dark mutation environment or low texture area. In order to improve the robustness and pose accuracy of the SLAM, a fast monocular visual–inertial odometry using point and line features is proposed in this paper. Firstly, the point and line features of the image are extracted for feature matching, and the IMU measurement data is extracted as the initialization state estimation information of the sliding window. The following IMU measurements are pre-integrated to constrain the successive IMU body states during optimization. Secondly, the proposed VIO uses geometric constraints to delete point-line matching outliers, and point-line feature endpoints are projected into a normalized plane for triangulation to build re-projection residuals. Finally, after obtaining the initial estimation of this system and the IMU body state predicted by IMU measurement, the sliding window optimization method is used to get the precise position of camera. The experiments are evaluated on public datasets (EuRoc MAV dataset and PennCOSYVIO dataset). The pose accuracy of this algorithm is more accurate than VINS-mono, and the speed is faster than PL-VIO.