A Visual Inertial SLAM Method based on Manhattan World assumption

• 1、School of Artificial Intelligence, University of Posts and Telecommunications,Beijing100876,China

Abstract：Aiming at the problem that the existing VI-SLAM algorithm lacks the observability of the IMU acceleration when the mobile robot is in a state of slow or uniform motion, and the problem of the reduced localization accuracy in the environment of illumination transformation and low texture. This method is a point line fusion VI-SLAM algorithm, and the Manhattan world assumption is introduced into Visual-SLAM problem. It can not only effectively improve the positioning accuracy of the Visual-SLAM system in structured scenes, but also detect the positioning state of the mobile robot in real timeA Visual Inertial SLAM Method based on Manhattan World assumption, and improve the robustness of the Visual-SLAM system; At the same time, a Manhattan World assumption pose constraint construction method based on local map is designed at Manhattan-SLAM back-end pose graph optimization, which effectively reduces redundant Manhattan World assumption pose constraints. The experimental results based on EuRoC dataset show that, by introducing the Manhattan World assumption, Manhattan-SLAM algorithm proposed in this paper has better performance in structured scenes than the mainstream VI-SLAM algorithm.

Keywords： SLAM Manhattan World assumption visual-inertial fusion mobile robot