PSL-SLAM: a monocular SLAM system using points and structure lines in Manhattan World

The performance of feature matching algorithms is well known to be one of the main Achilles heels of visual SLAM algorithms, and particularly for point-based visual SLAM. Which is prone to fail in low-textured ,enarios like man-made environments where points are insufficient. Yet, many environments in which, despite being low textured, can still reliably estimate line-based geometric primitives. The line-based structural features in the Manhattan world encode useful geometric information of parallelism, orthogonality, and coplanarity in the scene. By fully exploiting these structural features, we propose a novel monocular SLAM system merging feature points and structure lines, which can provide a more accurate estimation of camera poses. To integrate the structure lines into the framework of the system, we have made efforts in the detection, parameterization, feature fusion, and optimization modules of the structure lines. First, we used the consistency of the direction of the structure lines and the vanishing points to extract the structure lines. In the optimization module, we incorporated the error model of the structure lines into the nonlinear optimization framework, and proposed a new optimization strategy. Finally, a complete SLAM system based on points and structure lines is designed. With structure lines as a new observation, the robustness of the matching algorithm between consecutive frames in low-texture scenes is increased, ensuring continuous updating of the tracking thread when the feature points are lost. Secondly, the dominant directions of structure lines can provide global effective constraints to reduce the accumulated orientation errors and the position drift in consequence. Experiments in man-made environments have demonstrated that the proposed system outperforms existing state-of-the-art monocular SLAM systems in terms of accuracy and robustness.