Toward Consistent and Efficient Map-Based Visual-Inertial Localization: Theory Framework and Filter Design

This article focuses on designing a consistent and efficient filter for visual-inertial localization given prebuilt map. First, we propose new Lie group with its algebra based which novel invariant extended Kalman (invariant EKF) is designed. We theoretically prove that, when do not consider the uncertainty of map information, proposed EKF able to naturally preserve correct observability properties system. To introduce Schmidt filter. With filter, information can be taken into consideration avoid overconfident estimation while computation cost only increases linearly size keyframes. In addition, an easily implemented observability-constrained technique because directly combining cannot maintain system that considers information. Finally, validate our system's high consistency, accuracy, efficiency via extensive simulations real-world experiments.