Map-Based Visual-Inertial Localization: Consistency and Complexity

Drift-free localization is essential for autonomous vehicles. In this letter, we address the problem by proposing a filter-based framework, which integrates visual-inertial odometry and measurements from pre-built map. transformation between frame map augmented into system state vector estimated on fly. Besides, maintain keyframe poses employ Schmidt extended Kalman filter to update partially so that uncertainty of information can be consistently considered with low computational complexity. Moreover, theoretically demonstrate ever-changing linearization points make original four-dimensional unobservable subspace vanish, leading inconsistent estimation in practice. To relieve problem, first-estimate Jacobian (FEJ) technique correct observability properties system. Furthermore, introduce an observability-constrained updating method compensate significant accumulated error after long-term absence map-based measurements. Finally, evaluating through both simulation real-world experiments, confirm has good consistency