On the Consistency of Vision-Aided Inertial Navigation

Abstract

In this paper, we study estimator inconsistency in Vision-aided Inertial Navigation Systems (VINS). We show that standard (linearized) estimation approaches, such as the Extended Kalman Filter (EKF), can fundamentally alter the system observability properties, in terms of the number and structure of the unobservable directions. This in turn allows the influx of spurious information, leading to inconsistency. To address this issue, we propose an Observability-Constrained VINS (OC-VINS) methodology that explicitly adheres to the observability properties of the true system. We apply our approach to the Multi-State Constraint Kalman Filter (MSC-KF), and provide both simulation and experimental validation of the effectiveness of our method for improving estimator consistency.

Detecting and dealing with hovering maneuvers in vision-aided inertial navigation systems

“…Among these methods, the MultiState Constrained Kalman Filter (MSC-KF) [1] exploits all available geometric information provided by the camera measurements, while keeping its computational complexity linear in the number of features observed over the filter's window. Although the MSC-KF has been successfully applied to various applications (e.g., [1], [5]), and has been demonstrated to operate in real time [15], [16], it is not suitable for scenarios that include hovering over the same scene, since it requires sufficient baseline between the camera poses within the sliding window.…”

Exploiting Urban Scenes for Vision-aided Inertial Navigation

“…• The Observability-Constraint OC-MSC-KF of [8] using only observations of points (referred to as OC-VINS w/o lines in Fig. 3).…”

“…3). As explained in [8], due to estimation errors, the linearized error-state model of the MSC-KF erroneously perceives rotations about the gravity as observable, thus leading to inconsistency. The OC-MSC-KF addresses this problem by appropriately modifying the corresponding Jacobians and ensuring that no information from the measurements is injected along the direction of rotations about gravity.…”

“…Due to space limitations, we omit the description of the measurement model and the corresponding observability matrix for point features, since this has been investigated thoroughly in existing literature (see [15,8] and references therein). Instead we hereafter focus on the measurement model, corresponding to observations of lines of known direction.…”

“…Point features can be found in most natural and man-made environments, and as shown in [7,10,8], they provide sufficient information for estimating all degrees of freedom (d.o.f.) of a VINS, except the global position and yaw.…”