RD-VIO: Robust Visual-Inertial Odometry for Mobile Augmented Reality in Dynamic Environments (2310.15072v3)

Abstract: It is typically challenging for visual or visual-inertial odometry systems to handle the problems of dynamic scenes and pure rotation. In this work, we design a novel visual-inertial odometry (VIO) system called RD-VIO to handle both of these two problems. Firstly, we propose an IMU-PARSAC algorithm which can robustly detect and match keypoints in a two-stage process. In the first state, landmarks are matched with new keypoints using visual and IMU measurements. We collect statistical information from the matching and then guide the intra-keypoint matching in the second stage. Secondly, to handle the problem of pure rotation, we detect the motion type and adapt the deferred-triangulation technique during the data-association process. We make the pure-rotational frames into the special subframes. When solving the visual-inertial bundle adjustment, they provide additional constraints to the pure-rotational motion. We evaluate the proposed VIO system on public datasets and online comparison. Experiments show the proposed RD-VIO has obvious advantages over other methods in dynamic environments. The source code is available at: \href{https://github.com/openxrlab/xrslam}{{\fontfamily{pcr}\selectfont https://github.com/openxrlab/xrslam}}.

• The paper introduces the IMU-PARSAC algorithm that robustly removes dynamic keypoints by integrating visual and inertial data.
• The paper incorporates a novel subframe strategy to reduce drift and enhance stability during pure rotational motions.
• The paper demonstrates superior performance against state-of-the-art VIO systems on EuRoC and ADVIO datasets, promising reliable mobile AR experiences.

Overview of RD-VIO: Robust Visual-Inertial Odometry for Mobile AR in Dynamic Environments

The paper "RD-VIO: Robust Visual-Inertial Odometry for Mobile Augmented Reality in Dynamic Environments" presents a novel approach to visual-inertial odometry (VIO) systems, addressing specific challenges related to dynamic scenes and pure rotational motions. The authors introduce RD-VIO, an advanced VIO system, with significant contributions in keypoint detection and handling degenerate motion cases.

Key Contributions

The primary contributions of the paper are twofold:

1. IMU-PARSAC Algorithm: The authors introduce a robust method for detecting and removing dynamic keypoints using an iterative scheme that leverages both IMU and visual measurements. This two-stage process refines the accuracy of landmark tracking amidst moving objects, outperforming existing dynamic scene handling strategies like RANSAC.
2. Subframe Strategy for Pure Rotations: The paper introduces an innovative subframe strategy into the sliding window optimization process, significantly improving the system's robustness under conditions of pure rotation. By utilizing deferred-triangulation and treating pure-rotational frames as special subframes, RD-VIO reduces drift and enhances stability.

Evaluation and Results

The system's performance was rigorously evaluated on the EuRoC and ADVIO datasets, along with comparative assessments against other state-of-the-art VIO systems such as VINS-Fusion and ORB-SLAM3. The results demonstrate RD-VIO’s superior accuracy and robustness, particularly in dynamic environments and during degenerate motion. Specifically:

• Dynamic Environments: RD-VIO exhibits strong robustness in dynamic scenes by effectively handling moving objects and minimizing the negative impact of such disturbances on tracking accuracy.
• Pure Rotational Motions: The subframe strategy demonstrates a marked improvement in stability during pure rotational motions, as evidenced by reduced drift and enhanced tracking precision.

Implications and Future Work

The development of RD-VIO provides significant implications for both theoretical and practical applications in augmented reality (AR). The robust handling of dynamic scenes and pure rotations offers a path forward for reliable AR experiences on mobile devices. The released source code allows for community engagement and further innovation in mobile AR applications.

However, while RD-VIO effectively addresses many current challenges, the authors acknowledge limitations in situations with prolonged absence of valid visual observations. Future research could explore integrations with alternative odometry methods, such as pure inertial or wireless tracking, to further bolster performance in these edge cases.

In summary, RD-VIO represents a notable advancement in the field of visual-inertial odometry, offering practical solutions to persistent challenges in dynamic and degenerate environments. Its adoption and adaptation hold potential for further improvements in mobile AR technology.