Visual SLAM based on Collaborative Optimization of Points, Lines, and Planes

Abstract

A crucial step in implementing adaptive navigation for mobile robots is simultaneous localization and mapping (SLAM). While visual SLAM based on point features excels in pose estimation, its positioning accuracy can significantly decrease or even fail in indoor environments with sparse textures and significant lighting changes. To address this issue, this paper proposes a visual SLAM algorithm that combines point, line, and plane features for collaborative optimization under the constraints of the Manhattan world coordinate system. The algorithm begins by extracting point, line, and plane features, uti-lizing the orthogonal and parallel relationships of planar features in indoor environments to establish the dominant direction of the Manhattan coordinate system. Constrained by the Manhattan World Hypothesis, a pose decoupling approach is then employed to optimize and gradually minimize drift errors. Experimental results demonstrate that the visual SLAM system can achieve more accurate positioning in low-texture indoor environments.