Journal of Frontiers of Computer Science and Technology ›› 2021, Vol. 15 ›› Issue (8): 1546-1554.DOI: 10.3778/j.issn.1673-9418.2005043

• Theory and Algorithm • Previous Articles     Next Articles

Research on Initialization Algorithm for Visual-Inertial SLAM System

LIU Gang, GE Hongwei   

  1. 1. Jiangsu Provincial Engineering Laboratory of Pattern Recognition and Computational Intelligence, Jiangnan University, Wuxi, Jiangsu 214122, China
    2. School of Artificial Intelligence and Computer Science, Jiangnan University, Wuxi, Jiangsu 214122, China
  • Online:2021-08-01 Published:2021-08-02

视觉惯导SLAM初始化算法研究

刘刚葛洪伟   

  1. 1. 江苏省模式识别与计算智能工程实验室(江南大学),江苏 无锡 214122
    2. 江南大学 人工智能与计算机学院,江苏 无锡 214122

Abstract:

Monocular vision and inertial simultaneous localization and mapping (SLAM) system is becoming more and more popular in practical engineering applications because it can achieve the complementarity in use scenarios and lower hardware cost. Recent results have shown that optimization-based fusion strategies outperform filtering strategies. However, the optimization-based SLAM algorithm of vision inertial navigation fusion is highly nonlinear, and its performance highly depends on the accuracy of the estimation of the initial parameters of the system state. The inertial measurement unit needs acceleration excitation, which means that it cannot start from the static state, but must start from the unknown motion state. Therefore, accurate estimation of the initial state is the key to the high robustness of the algorithm and the first step of the vision inertial fusion algorithm. By analyzing the pre integration algorithm of inertial measurement unit, an initialization estimation system based on convex optimization is derived, and the initial states are solved jointly considering the constraints of the gravity acceleration. More importantly, a novel method is proposed to determine the termination condition of the initialization algorithm by measuring the estimation effect with Fisher information, which improves the accuracy of the algorithm and shortens the initialization time. Experiments on Euroc dataset show that the new algorithm has a more precise and robust initial state.

Key words: simultaneous localization and mapping (SLAM), visual inertial alignment, preintegration, inertial navigation, Fisher information

摘要:

单目视觉系统融合惯性测量单元的同时定位与地图构建(SLAM)系统,能实现在使用场景上的互补,以及较低的硬件成本,在实际工程应用中越来越受到青睐。最近的研究表明,基于优化的SLAM算法性能优于基于滤波的SLAM算法。基于优化的视觉惯导融合SLAM算法具有高度非线性化的特点,其性能高度依赖于系统初始状态估计的准确性;惯性测量单元需要加速度激励,这意味着不能从静止状态启动,而必须从未知的运动状态启动,因此如何确定这一未知的初始状态显得尤为重要。综上可知,对初始状态准确的估计是SLAM算法具有高精确鲁棒性的关键,也是视觉惯性融合算法的第一步。通过对惯性测量单元预积分算法的分析,推导出一种凸优化的初始化估计系统。在综合考虑了重力加速度的约束条件下,对各初始状态进行联合求解。更重要的是,提出了一种新颖的方法,即通过费歇尔信息衡量估计效果的好坏来确定初始化算法的终止条件,提高算法精确度的同时也缩短了初始化的时间。在Euroc数据集上的实验表明,该算法具有更高精确鲁棒的初始状态。

关键词: 同时定位与地图构建(SLAM), 视觉惯性对准, 预积分, 惯性导航, 费歇尔信息