**5. Conclusions**

In this paper, we realized a novel approach for online multi-sensor calibration implemented using a voxel-based CNN and 3D convolutional kernels. Our method aims to calibrate between sensors that can measure 3D space. In particular, the voxelization that converts the input 3D point cloud into voxel and the AM introduced to find the correlation of features between the reference and target sensors contributed greatly to the completeness of the proposed method. We demonstrated through experiments that the proposed method can perform both LiDAR-stereo camera calibration and LiDAR-LiDAR calibration. In the calibration of the LiDAR-stereo camera combination, the proposed method showed experimental results that surpassed all existing CNN-based calibration methods for the LiDAR-camera combination. We demonstrated the effects of iterative refinement on the five networks and the effects of temporal filtering through experiments. The proposed method achieved a rotation error of less than 0.1◦ and a translation error of less than 1 cm on both the KITTI and Oxford datasets.

**Author Contributions:** Conceptualization, J.S. and J.L.; methodology, J.S. and J.L.; software, J.S.; validation, J.S. and J.L.; formal analysis, J.S.; data curation, J.S.; writing—original draft preparation, J.S.; writing—review and editing, J.S. and J.L.; visualization, J.S.; supervision, J.L.; project administration, J.L.; funding acquisition, J.L. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was supported by Basic Science Research Program through the National Research Foundation of Korea (NRF), funded by the Ministry of Education (NRF-2016R1D1A1B02014422).

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Conflicts of Interest:** The authors declare no conflict of interest.
