Overall, the results indicate that combining efficient point-based and voxel-based features provides a computationally efficient framework for MLS point cloud understanding under the evaluated experimental conditions.
Abstract
Mobile LiDAR Systems (MLS) provide rich three-dimensional geometric information essential for autonomous driving, urban mapping, and smart city applications. However, accurate understanding of MLS point clouds remains challenging due to irregular sampling, varying density, occlusion, viewpoint changes, and the high computational cost of large-scale 3D data processing. This paper presents a lightweight two-stage hybrid framework for segmentation and classification of MLS LiDAR data. In the first stage, a dual-branch network combines point-wise local feature extraction with low-resolution voxel-based contextual encoding to generate reliable object candidates while maintaining moderate memory consumption. In the second stage, an adaptive cylindrical voxelization strategy, designed to align with the radial scanning geometry of LiDAR sensors, is applied to each candidate. Rotation-invariant geometric descriptors derived from Local Reference Frame (LRF) and Point Pair Features (PPF) are further fused with the learned descriptors to improve robustness under orientation and viewpoint variations. Experimental evaluations on the KITTI benchmark demonstrate that the proposed method achieves 93.8% average classification accuracy and 89.1% overall segmentation accuracy, while requiring, on average, only 8.7 GB of VRAM and 69ms inference time per scene. The proposed method achieves competitive performance relative to representative point-based, voxel-based, and hybrid baselines. Cross-dataset evaluation on a merged MLS dataset provides an initial indication of performance on unseen data; however, the limited and imbalanced nature of this dataset does not permit a strong claim regarding generalization capability. Ablation studies further verify the effectiveness of dual-branch feature fusion and adaptive cylindrical voxelization. Overall, the results indicate that combining efficient point-based and voxel-based features provides a computationally efficient framework for MLS point cloud understanding under the evaluated experimental conditions.
LiDAR point cloud semantic segmentation is a fundamental component of 3D perception for autonomous driving and robotics. Projection-based methods map 3D point clouds into 2D representations for feature extraction using 2D networks. However, aggregating multiple points during projection can obscure geometric differences...
Raw multi-line LiDAR data in geospatial surveying are characterized by complex preprocessing, substantial redundancy, and limited efficiency in hierarchical object extraction. This paper presents a mapping-oriented framework for adaptive voxelization and hierarchical object extraction directly from raw LiDAR network pa...
Meng Li, Xiao-Han Gao, Quan-Ying Li et al.· International Conference on...· 0 citations
3D object detection from LiDAR point clouds is a fundamental perception task in autonomous driving, aiming to accurately localize and classify surrounding objects in sparse and irregular point-cloud data. CenterPoint is a widely adopted anchor-free baseline for LiDAR-based 3D object detection, but it often suffers from...
Jia-Wei Lin· The International Conference...· 0 citations
3D object detection is a core task in intelligent perception, and LiDAR point clouds provide precise spatial geometric and scale information. To address issues such as high sparsity, non-uniform distribution, and insufficient feature representation of distant objects in LiDAR point clouds, this paper proposes DLP-Pilla...
Industrial environments often contain dynamic or temporary objects, such as mobile containers, trash bins, workers, and forklifts, which may degrade LiDAR-based maps by introducing transient structures into the reconstructed representation of the environment. This work investigates the removal of objects of interest fr...
Kennedy O. S. Mota, Luís Garrote, A. Mendes et al.· International Conference on...· 0 citations
This paper studies LiDAR-assisted RGB semantic segmentation, where LiDAR point-cloud features improve dense pixel-wise predictions in the RGB image plane rather than 3D point labels or detections. Existing 2D fusion methods often project point clouds into shallow depth/range maps and apply spatially uniform fusion, dis...
R. Lin, Peng-Cheng Weng, Ge Wang et al.· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.