The content discusses a spatiotemporal hand-eye calibration algorithm designed for accurately aligning the estimated trajectory from visual(-inertial) odometry (VO/VIO) with a ground-truth trajectory obtained from a high-precision system like a motion capture (MoCap) system.
The key highlights are:
The algorithm addresses two main challenges in trajectory alignment: non-corresponding timestamps and different reference frames between the VO/VIO and ground-truth trajectories.
For time alignment, the algorithm improves the correlation analysis of the screw invariant to obtain synchronized trajectories with higher precision.
For spatial calibration, the algorithm constructs linear equations using local relative poses based on rotational constraint to fully utilize the motion information, rather than using global or inter-frame strategies.
A robust kernel based on screw theory is introduced to stabilize the linear solution, and a RANSAC framework is used to recover inlier data.
A nonlinear optimization tool is designed to jointly refine the time offset and the linear extrinsic solution.
The proposed algorithm demonstrates improved accuracy and robustness compared to state-of-the-art methods, especially when handling noisy and drifting VO/VIO trajectories. Experiments on public and simulated datasets, as well as the authors' own dataset collected using a VR headset and a MoCap system, validate the effectiveness of the algorithm.
לשפה אחרת
מתוכן המקור
arxiv.org
תובנות מפתח מזוקקות מ:
by Zichao Shu,L... ב- arxiv.org 04-24-2024
https://arxiv.org/pdf/2404.14894.pdfשאלות מעמיקות