.png){kind=icon}
.png){kind=icon}
Return
Multi-LiDAR Registration: A Joint Sensor-Centric Optimization Approach
.png){kind=icon}
PRE
AI
DOI：10.1109/LSENS.2025.3644412.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Multi-light detection and ranging (LiDAR) fusion is widely used to increase scene coverage and improve 3-D reconstruction quality, but jointly registering point sets from scanners with different resolutions, scales, fields of view, and noise characteristics remains difficult and directly impacts sensing accuracy. This letter presents a sensor-centric multi-LiDAR joint registration framework, which includes the following. First, it lifts iterative closest points (ICP) to a high-dimensional formulation to jointly align multiple scans with block-structured rotation couplings. Second, it introduces data-driven weighting and a two-stage outlier diagnostics procedure tailored to cross-sensor inconsistencies. Lastly, it performs uncertainty-aware regularization using closed-form covariances for both rotation and translation. The method preserves simple singular value decomposition (SVD)-based updates while explicitly addressing heterogeneous sensor characteristics. Validation on two hardware platforms-a dual 2-D spinning setup (SICK TIM520 and Hokuyo UST10LX) and a dual Ouster OS1128 suite-demonstrates sensor-system-level accuracy gains, reducing accumulated pose error by 26.9%--40.8% relative to representative ICP variants, with run times approximate to 3x faster than point-to-plane ICP and approximate to 38x faster than Go-ICP (slightly slower than efficient sparse ICP). These results substantiate a direct contribution to sensor systems by improving multi-LiDAR integration robustness, accuracy, and deployment practicality.
Keywords:
Robot sensing systems
Noise
Measurement
Accuracy
Uncertainty
Translation
Robustness
Laser radar
Three-dimensional displays
Optimization
Sensor applications
motion estimation
navigation
point-cloud registration
rigid transformation
robotic perception
Journal
I
IF:
2.2
Papers:
354
Citations:
3.1K