.png){kind=icon}
.png){kind=icon}
Return
M<inline-formula><tex-math notation="LaTeX">$^{\mathbf{2}}$</tex-math></inline-formula>-Calibr: Targetless Spatiotemporal Calibration for Multisensor via Multioutput Gaussian Processes
.png){kind=icon}
PRE
AI
F
J
W
B
B
DOI：10.1109/TRO.2026.3699427.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Multisensor fusion has been widely accepted as the mainstream approach for autonomous driving and robotic tasks, which is founded on precise spatiotemporal calibration. However, significant differences in sampling frequency and modality among various sensors, such as LiDAR, IMU, and cameras, present considerable challenges to calibration, especially under severe motion conditions. To this end, we propose M<inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$^{\mathbf{2}}$</tex-math></inline-formula>-Calibr, a targetless spatiotemporal calibration for multisensor systems via a multioutput Gaussian process (MOGP). First, continuous-time motion states are modeled via MOGP regression. Building on this, Point-MOGP is proposed to refine per-point LiDAR motion states, reducing motion-induced errors and producing an undistorted point-cloud map. This map is then utilized to recover the absolute scale of camera motion, further enhancing calibration accuracy. Finally, to further enhance the calibration performance under rapid motion conditions, a trajectory correlation analysis method incorporating dynamic time warping (DTW) is proposed, which computes the minimal distortion path and the maximum trajectory correlation between motion timestamps, enabling fast and accurate estimation of temporal offsets. Extensive experiments on benchmark and custom datasets demonstrate that M<inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$^{\mathbf{2}}$</tex-math></inline-formula>-Calibr performs well in accuracy.
Keywords:
Multioutput Gaussian processes (MOGP)
multisensor
spatiotemporal calibration
undistorted pointcloud
Journal
IF:
10.5
Papers:
3.3K
Citations:
2.8W
Organization
Cited Papers