Investigating a Unified 3-D Object Detection Method for Different Multibeam LiDAR
Tao, Ziming Mao, Jianxu Wang, Yaonan Liu, Caiping Yi, Junfei He, Zhenyu Chang, Xiaojun Zhang, Hui
Tao, Ziming
Mao, Jianxu
Wang, Yaonan
Liu, Caiping
Yi, Junfei
He, Zhenyu
Chang, Xiaojun
Zhang, Hui
Supervisor
Department
Computer Vision
Embargo End Date
Type
Journal article
Date
2025
License
Language
English
Collections
Research Projects
Organizational Units
Journal Issue
Abstract
Over the past few years, interest in improving the perception performance of LiDAR has increased in the autonomous driving. Current approaches primarily optimize models for specific LiDAR beam configurations, overlooking cross-beam compatibility, which severely restricts their applicability to datasets from different multibeam sensors. Our observations suggest that existing methods cannot effectively leverage open-source datasets to enhance custom dataset training, primarily due to significant configuration discrepancies, especially in LiDAR beam characteristics. This issue prompted us to design a unified detector that can support multibeam datasets and fully exploit the advantages of open-source data to supplement custom datasets. This article proposes UniDMB, a novel 3-D object detection method for heterogeneous multibeam LiDAR systems. To advance multibeam LiDAR research, we additionally introduce a new simulated dataset covering various beam configurations. Specifically, UniDMB mainly includes an alignment module and a global state space module. The alignment module aligns different point clouds at the spatial and feature levels, eliminating differences in data structures, and the global state space module has a larger receptive field, which is helpful for recognizing scene layouts. Furthermore, dataset-specific heads are used for prediction. In addition, we construct a dataset via simulations that consist of point clouds from different multibeam LiDAR sensors, e.g., 16- and 64-beam sensors. Finally, we conduct extensive experiments on simulation datasets to demonstrate the superiority of UniDMB, and the validity of this method is verified on the basis of the KITTI open-source dataset and the simulation dataset.
Citation
Z. Tao et al., "Investigating a Unified 3-D Object Detection Method for Different Multibeam LiDAR," in IEEE Transactions on Industrial Informatics, doi: 10.1109/TII.2025.3609110
Source
IEEE Transactions on Industrial Informatics
Conference
Keywords
3-D Object Detection, Multibeam LiDAR, Point Cloud, Unified 3-D Detector
Subjects
Source
Publisher
IEEE