CaLiV: LiDAR-to-Vehicle Calibration of Arbitrary Sensor Setups

TL;DR

CaLiV tackles extrinsic calibration for multi-LiDAR systems with non-overlapping FoVs without external sensing devices. It introduces a target-based framework that first solves Sensor-to-Sensor calibration and then derives Sensor-to-Vehicle calibration by grounding both in a shared calibration frame via GMMCalib, using motion to create necessary FoV overlap and an Unscented Kalman Filter for vehicle poses. The approach achieves high accuracy in both simulation and real-world experiments, outperforming existing non-overlapping S2S and S2V methods, and is released as open-source. This work enables robust, environment-independent calibration for complex sensor configurations, improving safety-critical perception and planning in autonomous systems.

Abstract

In autonomous systems, sensor calibration is essential for safe and efficient navigation in dynamic environments. Accurate calibration is a prerequisite for reliable perception and planning tasks such as object detection and obstacle avoidance. Many existing LiDAR calibration methods require overlapping fields of view, while others use external sensing devices or postulate a feature-rich environment. In addition, Sensor-to-Vehicle calibration is not supported by the vast majority of calibration algorithms. In this work, we propose a novel target-based technique for extrinsic Sensor-to-Sensor and Sensor-to-Vehicle calibration of multi-LiDAR systems called CaLiV. This algorithm works for non-overlapping fields of view and does not require any external sensing devices. First, we apply motion to produce field of view overlaps and utilize a simple Unscented Kalman Filter to obtain vehicle poses. Then, we use the Gaussian mixture model-based registration framework GMMCalib to align the point clouds in a common calibration frame. Finally, we reduce the task of recovering the sensor extrinsics to a minimization problem. We show that both translational and rotational Sensor-to-Sensor errors can be solved accurately by our method. In addition, all Sensor-to-Vehicle rotation angles can also be calibrated with high accuracy. We validate the simulation results in real-world experiments. The code is open-source and available on https://github.com/TUMFTM/CaLiV.

Figures (7)

• Figure 1: CaLiV: Data is collected by driving a curved trajectory. The data is first aligned using vehicle poses and different perspectives of the calibration target. The S2S calibration is first computed by the algorithm in the optimization phase, and it subsequently serves as the basis for the S2V optimization. The input to the S2S and S2V optimization is a rough point cloud alignment and the shape of the reconstructed target, both generated using GMMCalib GMMCalib.
• Figure 2: Spatial relationship between the frames $\{V^i\}$ of a moving system at a specific time steps $t_i$ with two LiDAR frames $\{L^i_1\}$ and $\{L^i_2\}$, a global reference frame $\{R\}$, and a common calibration frame $\{C\}$.
• Figure 3: This figure illustrates the optimization process of the calibration. The two transformation estimates are used to transform the point sets represented by the bounding box of the calibration target shape.
• Figure 4: Left: Curved maneuver with the test vehicle in the real-world setup. Right: A chair as a calibration target and the cubic validation target.
• Figure 5: The distribution of the non-overlapping S2S translation and rotation errors of CaLiV.