A Square Root Inverse Filter for Efficient Vision-aided Inertial Navigation on Mobile Devices

Abstract

Abstract-In this paper, we present a square-root inverse sliding window filter (SR-ISWF) for vision-aided inertial navigation systems (VINS). While regular inverse filters suffer from numerical issues, employing their square-root equivalent enables the usage of single-precision number representations, thus achieving considerable speed ups as compared to doubleprecision alternatives on resource-constrained mobile platforms. Besides a detailed description of the SR-ISWF for VINS, which focuses on the numerical procedures that enable exploiting the problem's structure for gaining in efficiency, this paper presents a thorough validation of the algorithm's processing requirements and achieved accuracy. In particular, experiments are conducted using a commercial-grade cell phone, where the proposed algorithm is shown to achieve the same level of estimation accuracy, when compared to state-of-the-art VINS algorithms, with significantly higher speed.

Visual-Inertial Navigation: A Concise Review

“…In particular, by exploiting the observability-based methodology proposed in our prior work [46,47,48,49], different observabilityconstrained (OC)-MSCKF algorithms have been developed to improve the filter consistency by enforcing the correct observability properties of the linearized VINS [19,38,39,40,50,51,52]. A square-root inverse version of the MSCKF, i.e., the square-root inverse sliding window filter (SR-ISWF) [6,53] was introduced to improve the computational efficiency and numerical stability to enable VINS running on mobile devices with limited resources while not sacrificing estimation accuracy. We have introduced the optimal state constraint (OSC)-EKF [54,55] that first optimally extracts all the information contained in the visual measurements about the relative camera poses in a sliding window and then uses these inferred relative-pose measurements in the EKF update.…”

“…Most INS rely on a 6-axis inertial measurement unit (IMU) that measures the local linear acceleration and angular velocity of the platform to which it is rigidly connected. With the recent advancements of hardware design and manufacturing, low-cost light-weight micro-electro-mechanical (MEMS) IMUs have become ubiquitous [3,4,5], which enables high-accuracy localization for, among others, mobile devices [6] and micro aerial vehicles (MAVs) [7,8,9,10,11], holding huge implications in a wide range of emerging applications from mobile augmented reality (AR) [12,13] and virtual reality (VR) [14] to autonomous driving [15,16]. Unfortunately, simple integration of high-rate IMU measurements that are corrupted by noise and bias, often results in pose estimates unreliable for long-term navigation.…”

An Efficient Schmidt-EKF for 3D Visual-Inertial SLAM

“…While SLAM estimators -by jointly estimating the location of the sensor platform and the features in the surrounding environment -are able to easily incorporate loop closure constraints to bound localization errors and have attracted much research attention in the past three decades [8,1,6,3], there are also significant research efforts devoted to open-loop VIO systems (e.g., [30,13,14,22,17,50,37,49,53,5,2,15,40]). For example, a hybrid MSCKF/SLAM estimator was developed for VIO [23], which retains features that can be continuously tracked beyond the sliding window in the state as SLAM features while removing them when they get lost.…”

“…This localization solution has the advantages of being both cheap and ubiquitous, and has the potential to provide position and orientation (pose) estimates which are on-par in terms of accuracy with more expensive sensors such as LiDAR. To date, various algorithms are available for VINS problems including visual-inertial (VI)-SLAM [19,45] and visual-inertial odometry (VIO) [30,29,22], such as the extended Kalman filter (EKF) [30,20,14,22,17,16,50,37], unscented Kalman filter (UKF) [10,4], and batch or slidingwindow optimization methods [46,18,21,33,52,45,40], among which the EKF-based approaches remain arguably the most popular for resource constrained devices because of their efficiency. While current approaches can perform well over a short period of time in a small-scale environment (e.g., see [13,22,15]), they are not robust and accurate enough for long-term, large-scale deployments in challenging environments, due to their limited available resources of sensing, memory and computation, which, if not properly addressed, often result in short mission duration or intractable real-time estimator performance.…”

Communication constrained cloud-based long-term visual localization in real time

“…Generally, there are two branches of methods in this area. The first branch utilizes a nonlinear filter to estimate the pose, which is very efficient and light-weighted, thus is suitable for mobile platform with limited computational resources [10], [11]. Another branch of methods leverages non-linear optimization techniques based on local keyframes, i.e.…”