GMMCalib: Extrinsic Calibration of LiDAR Sensors using GMM-based Joint Registration

TL;DR

GMMCalib tackles robust extrinsic LiDAR calibration by replacing pairwise ICP with a probabilistic joint-registration approach based on a Gaussian Mixture Model. The method jointly estimates per-sensor rigid transforms while fitting a shared GMM to multiple observations, enabling robust calibration without fixing a reference cloud. Through simulation in CARLA and real-world EDGAR experiments, the study shows improved robustness and fewer miscalibrations compared to ICP variants, with a plausible target-based reconstruction as a validity check. The work highlights practical trade-offs, notably computational cost, and suggests directions for extending the framework to non-overlapping fields of view and integrating geometric priors for enhanced physical consistency.

Abstract

State-of-the-art LiDAR calibration frameworks mainly use non-probabilistic registration methods such as Iterative Closest Point (ICP) and its variants. These methods suffer from biased results due to their pair-wise registration procedure as well as their sensitivity to initialization and parameterization. This often leads to misalignments in the calibration process. Probabilistic registration methods compensate for these drawbacks by specifically modeling the probabilistic nature of the observations. This paper presents GMMCalib, an automatic target-based extrinsic calibration approach for multi-LiDAR systems. Using an implementation of a Gaussian Mixture Model (GMM)-based registration method that allows joint registration of multiple point clouds, this data-driven approach is compared to ICP algorithms. We perform simulation experiments using the digital twin of the EDGAR research vehicle and validate the results in a real-world environment. We also address the local minima problem of local registration methods for extrinsic sensor calibration and use a distance-based metric to evaluate the calibration results. Our results show that an increase in robustness against sensor miscalibrations can be achieved by using GMM-based registration algorithms. The code is open source and available on GitHub.

Figures (10)

• Figure 1: This figure illustrates the front (upper left), side (upper right) and pitched (lower left) view of the reconstructed calibration target colored in black and the ground truth point cloud (lower right) colored in blue.
• Figure 2: Spatial relationship between the LiDAR sensor frames and the calibrated frame with $\{L_1\}$ and $\{L_2\}$ representing the sensor coordinate frames and $\{R\}$ representing the calibration frame.
• Figure 3: A CARLA simulation environment showing three cubic calibration targets with edges of 0.5 meters in close proximity. The digital twin of the EDGAR edgar research vehicle is placed at the origin of the world.
• Figure 4: Real-world experimental setup with three cubic calibration targets with edges of 0.5 m placed on a regular road. Another target is placed at a distance of about 16 m.
• Figure 5: Spatial relationship between the LiDAR sensor frames and $\{\widetilde{L}_1\}$ with $\{L1\}$ and $\{L2\}$ representing the sensor frames and $\{\widetilde{L}_1\}$ represents the erroneously transformed frame.