VINS-Mono: A Robust and Versatile Monocular Visual-
Inertial State Estimator
Shaozu Cao, Jie Pan, Jieqi Shi and Shaojie Shen
Hong Kong University of Science and Technology
{shaozu.cao, jpanai, jshias, eeshaojie}@connect.ust.hk
I. Original Framework
Our system is based on the VINS-MONO[1] framework illustrated in the picture.
We first extract features from images provided by a camera sensor and operate
optical flow algorithms to complete a pre-treatment. Meanwhile, we do IMU pre-
integration between two consecutive frames. After pre-processing everything we need,
we perform a robust estimator initialization.
In the initialization procedure, we make use of visual structure from motion
together with visual-inertial alignment. This loosely-coupled sensor fusion method
offers us initial values we need to build a monocular tightly-coupled visual-inertial
system.
To change from loosely to tightly coupled, we proceed with a sliding window-
based optimization system for high accuracy and robust state estimation. We also
employ marginalization to bound the computational complexity of our optimization-
based VIO. What’s more, a lightweight motion-only visual-inertial BA is carried out to
boost the estimation.
评论0