PGD-VIO: An Accurate Plane-Aided Visual-Inertial Odometry with Graph-Based Drift Suppression

TL;DR

PGD-VIO addresses long-term drift in RGB-D visual-inertial odometry by fusing depth-informed point features and plane landmarks within an extended Kalman filter, and by a graph-based drift suppression that detects repetitive plane structures to correct cumulative error without loop closures. The method introduces a comprehensive EKF state including points, planes, and calibration, and leverages depth to improve point triangulation while treating planes as direct measurements. A novel graph matching framework over plane patches enables drift detection and de-drift updates, resulting in accurate localization and a compact, consistent plane map on two public datasets, outperforming several point-based and plane-aware baselines. Limitations arise when planar structures are sparse or highly repetitive, pointing to future work on leveraging environmental regularities in plane association and EKF updates.

Abstract

Generally, high-level features provide more geometrical information compared to point features, which can be exploited to further constrain motions. Planes are commonplace in man-made environments, offering an active means to reduce drift, due to their extensive spatial and temporal observability. To make full use of planar information, we propose a novel visual-inertial odometry (VIO) using an RGBD camera and an inertial measurement unit (IMU), effectively integrating point and plane features in an extended Kalman filter (EKF) framework. Depth information of point features is leveraged to improve the accuracy of point triangulation, while plane features serve as direct observations added into the state vector. Notably, to benefit long-term navigation,a novel graph-based drift detection strategy is proposed to search overlapping and identical structures in the plane map so that the cumulative drift is suppressed subsequently. The experimental results on two public datasets demonstrate that our system outperforms state-of-the-art methods in localization accuracy and meanwhile generates a compact and consistent plane map, free of expensive global bundle adjustment and loop closing techniques.

Figures (6)

• Figure 1: Illustration of the proposed PGD-VIO on the CID-SIMS sequence Floor3_1. Attributed to the drift suppression strategy, the system can detect overlapping and identical configurations in the plane map and align them to cope with cumulative errors, resulting in an accurate trajectory and a more consistent plane map.
• Figure 2: Overview of the proposed PGD-VIO system.
• Figure 3: Distinctive relative positions of two plane patches, viewed from a common perpendicular direction to their normal vectors. In each figure, the yellow dot is the closest point on plane $i$ to plane $j$ and the corresponding line depicts the distance $d_{ij}$, the blue dot and the blue line measure the distance $d_{ji}$ from plane $j$ to plane $i$ oppositely, and the green line marks their overlapping region, which is negative implying the parallel distance in (d) and (f).
• Figure 4: Pipeline of the drift suppression.
• Figure 5: Comparative trajectories of the evaluated methods on the CID-SIMS dataset. For visualization, the first 500 frames are used to align the trajectories with the ground truth.