Robot Visual Inertial SLAM Algorithm Based on Point-Line Features

In weakly textured situation, the point feature SLAM system has a large localization error due to the failure to extract enough point features, and even the running trajectory is lost. To address such problems, an improved point-line feature monocular VINS-mono (IPLM-VINS) system is proposed. Line feature information is introduced to the front-end of the system. In order to address the issue of a large number of redundant short lines in the EDlines, we proposed a length suppression strategy. In the back-end sliding window optimization, line feature reprojection error is incorporated. Furthermore, the Huber kernel function is applied to IMU residuals, point-line feature residuals, and loop closure constraints to mitigate the impact of outliers on the optimization results. Comparative experiments were conducted on the public EuRoC dataset, and the positioning accuracy and real-time performance of IPLM-VINS were significantly improved compared to several state-of-the-art VIO systems. The results demonstrate that the positioning accuracy of IPLM-VINS system is improved by 22.9% on average compared with that of VINS-mono. Experiments are carried out in real weakly textured scenes, and the localization accuracy of the improved algorithm is increased by more than 30%.