基于模型选择的视觉惯导SLAM初始化算法

Initialization algorithm for visual-inertial SLAM based on model selection

单目视觉惯导的同时定位与建图(simultaneous localization and mapping,SLAM)初始化阶段是整个系统重要且脆弱的一部分,需要一个鲁棒的初始化过程以确保系统的可用性.针对初始化过程中容易出现位姿估计不准确而导致整个系统定位漂移的问题,本文在VINS-Mono的系统框架下,提出了一种基于模型选择的自动初始化方法,使SLAM系统可以根据不同场景自动选择合适的模型来完成联合初始化,得到系统初始值.在Euroc公开数据集的实验验证了该方法的有效性,实验结果表明改进后的VINS-Mono系统的精度和鲁棒性均得到了提高.

The initialization stage of monocular visual-inertial simultaneous localization and mapping(SLAM)is an important and fragile part of the entire system,which requires a robust initialization process to ensure the availability of the system.Aimed at the problem of inaccurate pose estimation during the initialization process that leads to the drift of the entire system,an automatic initialization method based on model selection under the VINS-Mono system framework is proposed in this paper,so that the appropriate model can be automatically selected by the SLAM system to complete the joint initialization and get the initial value of the system.The effectiveness of the method is verified through experiments on the Euroc public dataset,the experimental results show that the accuracy and robustness of the improved VINS-Mono system have been improved.