Multi-LiCa: A Motion and Targetless Multi LiDAR-to-LiDAR Calibration Framework (2501.11088v1)

Abstract: Today's autonomous vehicles rely on a multitude of sensors to perceive their environment. To improve the perception or create redundancy, the sensor's alignment relative to each other must be known. With Multi-LiCa, we present a novel approach for the alignment, e.g. calibration. We present an automatic motion- and targetless approach for the extrinsic multi LiDAR-to-LiDAR calibration without the need for additional sensor modalities or an initial transformation input. We propose a two-step process with feature-based matching for the coarse alignment and a GICP-based fine registration in combination with a cost-based matching strategy. Our approach can be applied to any number of sensors and positions if there is a partial overlap between the field of view of single sensors. We show that our pipeline is better generalized to different sensor setups and scenarios and is on par or better in calibration accuracy than existing approaches. The presented framework is integrated in ROS 2 but can also be used as a standalone application. To build upon our work, our source code is available at: https://github.com/TUMFTM/Multi_LiCa.

• The paper presents a calibration framework that eliminates the need for auxiliary targets or motion, enabling flexible alignment of multiple LiDAR sensors.
• It employs a two-step approach using FPFH for coarse registration and TEASER++ with GICP for precise refinement of point cloud transformations.
• Evaluation on diverse datasets demonstrated lower RMSE values and enhanced robustness over traditional methods in autonomous vehicle sensor setups.

Multi-LiCa: A Motion- and Targetless Multi-LiDAR-to-LiDAR Calibration Framework

The paper outlines a novel framework named Multi-LiCa for the extrinsic calibration of multiple LiDAR sensors. Unlike traditional approaches, Multi-LiCa operates without the need for auxiliary targets or additional sensor modalities, which are often relied upon for accurate calibration. This framework specifically addresses the calibration challenges necessitated by the diverse sensor configurations found in autonomous vehicles.

Methodology

Multi-LiCa is characterized by a two-step calibration approach. Initially, it leverages Fast Point Feature Histograms (FPFH) to achieve coarse alignment of point clouds across LiDAR sensors using feature-based matching. TEASER++ is then employed to refine the initial estimates for transformations between point clouds. This is followed by a fine registration phase utilizing the Generalized Iterative Closest Point (GICP) algorithm. The framework also incorporates a strategy to sequentially calibrate LiDARs even in scenarios lacking direct field-of-view overlap with a central LiDAR. This is achieved through point cloud merging strategies and dynamically adjusting target clouds to maximize the overlap and thereby improve calibration potential.

An additional component of this research is the proposed LiDAR-to-Ground calibration method, which can compute specific transformation components, such as roll, pitch, and vertical translation, relative to a flat ground assumption. This step aims at simplifying the calibration process, especially when a LiDAR is inclined or positioned non-traditionally relative to the ground plane.

Results

The efficacy of Multi-LiCa was evaluated using datasets from EDGAR—a research vehicle equipped with a unique angulated multi-LiDAR setup—and the HeLiPR dataset, featuring various LiDAR configurations, including mixed types and resolutions. Notably, the framework demonstrated significant calibration accuracy, maintaining greater robustness across scenes and sensor setups when compared to existing methodologies like CROON.

In terms of translational and rotational accuracy, Multi-LiCa showed lower root mean square error (RMSE) values on both the HeLiPR and EDGAR datasets. The computed transformations closely matched the ground truth benchmarks, thereby underscoring the framework’s robustness and precision in handling diverse environmental configurations and sensor orientations.

Discussion and Implications

The research provides a significant contribution to multi-sensor calibration techniques in autonomous vehicle systems. The motionless and targetless calibration capabilities mean that Multi-LiCa can be applied in a wider range of operational environments without the need for complex setups involving predefined targets. This flexibility enhances its utility across varied autonomous platforms, potentially reducing deployment and implementation time in real-world scenarios.

Moreover, the broad applicability of the calibration method, combined with the open-source availability of its implementation, supports further adaptation and refinement by the research community. Anticipated future developments in the field may focus on enhancing the computational efficiency of the framework, expanding its application to other sensor modalities, and refining calibration accuracy through more sophisticated optimization techniques.

Overall, Multi-LiCa represents a compelling advancement in sensor calibration practices, bridging critical gaps in robustness, generalizability, and automation in multi-LiDAR systems.