| 刘政,张中月,吴长悦,王鹤.基于PointNet++算法的道路设施双形态点云分类研究[J].唐山学院学报,2025,38(6):23-28 |
| 基于PointNet++算法的道路设施双形态点云分类研究 |
| Research on Dual-Morphological Point Cloud Classification of Road Facility Based on PointNet++ |
| 投稿时间：2025-02-27 |
| DOI：10.16160/j.cnki.tsxyxb.2025.06.005 |
| 中文关键词: PointNet++ 道路附属设施 点云分类 |
| 英文关键词: PointNet++ road ancillary facilities point cloud classification |
| 基金项目: |
|
| 摘要点击次数: 428 |
| 全文下载次数: 227 |
| 中文摘要: |
| 针对复杂道路场景中由点云相互遮挡、设施结构残缺以及残缺点云样本稀缺性所导致的分类难题,文章提出了基于PointNet++算法的道路附属设施点云分类方法。通过构建涵盖行道树、路灯、垃圾桶等典型设施的道路场景点云数据集,利用PointNet++算法的多尺度特征提取能力,采用完整点云数据训练模型探究其对完整及残缺点云的分类鲁棒性。结果表明:该模型对完整点云的分类准确率达到93.6%,对残缺点云的分类准确率平均达到90%以上,显著优于传统手工分类方法。该研究验证了PointNet++算法模型对点云缺损的强容错能力,可为智慧道路巡检系统的设施自动化识别与城市三维点云缺损修复提供技术支持。 |
| 英文摘要: |
| To overcome classification challenges in complex road scenes caused by mutual occlusion of point clouds, facility structural defects, and scarcity of incomplete point cloud samples, this paper proposes a point cloud classification method for road ancillary facility based on PointNet++. By constructing a point cloud dataset of road scenes encompassing typical facilities such as roadside trees, road lamps and trash bins, and leveraging the multi-scale feature extraction capabilities of PointNet++, the model is trained solely on complete point cloud data to investigate its robustness in classifying both complete and incomplete point clouds. The results show that the classification accuracy of complete point clouds reaches 93.6%, while that of incomplete point clouds exceeds 90%, significantly outperforming traditional manual feature methods. This demonstrates PointNet++’s strong fault tolerance for point cloud defects, providing technical support for automated facility recognition in smart road inspection systems and the repair of urban 3D point cloud defects. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|