A Research of Visual-Inertial Simultaneous Localization and Mapping

Abstract

Simultaneous localization and mapping is a popular research topic in robotics, which aims to model the environment during the motion of a subject equipped with specific sensors, without a priori information about the environment. The ability to functionally complement two external sensors, the camera and the inertial measurement unit (IMU), gives it an advantage in terms of robustness. Therefore visual-inertial system (VINS) has a wide range of applications in the field of localization and mapping, including mobile robots, Automated driving systems (ADSs) 0 and aircraft. This study provides an introduction to the implementation of VINS for front-end odometry, back-end optimization, Loop closure detection and map building. Key problems and solutions in VINS, such as pose estimation, back-end optimization, and loop closure detection are highlighted. The characteristics of representative VINS methods are summarized to highlight the advantages of different method implementations. The whole process of VI-SLAM implementation is systematically discussed and combined with classical implementation methods, which helps to master the system implementation process and understand the differences of different working principles. This paper also presents the future trends and research directions of VINS, which can be used as a brief guide for novices and experienced researchers in the field of SLAM.