| 引用本文: | 陈 玲, 许 钢, 伏娜娜, 胡志锋, 郑书展.融合边缘检测的 3D 点云语义分割方法研究(J/M/D/N,J:杂志，M：书，D：论文，N：报纸).期刊名称,2022，39（5）：1-9 |
| CHEN X. Adap tive slidingmode contr ol for discrete2ti me multi2inputmulti2 out put systems[ J ]. Aut omatica, 2006, 42(6): 4272-435 |
|
| 摘要: |
| 针对点云分割中分割目标不明确,边缘不清晰,全局特征与边缘特征未能有效融合等问题,提出 了一种融合边缘检测的 3D 点云语义分割算法。 首先,通过 3D 点云语义分割网络对点云数据进行初步提取 区域内的全局语义特征;然后,采用引入了注意力机制的语义边缘检测网络,能够更好地对点云数据中的物 体进行特征提取增强,抑制非边缘信息的产生,得到了具有丰富的语义信息的边缘特征;最后,通过融合模块 将属于同一物体的语义特征融合起来进行分割细化处理,使得分割目标更精确;此外,使用了双重语义损失 函数,使网络产生具有更好边界的语义分割结果。 通过搭建实验平台和使用 S3DIS 标准数据集进行测试,改 进后的算法在数据集上的平均交互比为 70. 21%,在精度上较 KPConv 语义分割算法有所提高。 实验结果表 明:该算法能够有效改善物体边界分割不清晰、边缘信息模糊等问题,总体分割性能良好。 |
| 关键词: 3D 点云 语义分割 语义边缘检测 特征融合 |
| DOI： |
| 分类号: |
| 基金项目: |
|
| Research on 3D Point Cloud Semantic Segmentation Method Fused with Edge Detection |
|
CHEN Ling, XU Gang, FU Na-na, HU Zhi-feng, ZHENG Shu-zhan
|
|
Key Laboratory of Detection Technology and Energy Saving Devices of Anhui Province, Anhui University of Technology, Anhui Wuhu 241000, China
|
| Abstract: |
| Aiming at the problems of unclear segmentation targets, unclear edges, and ineffective fusion of global features and edge features in point cloud segmentation, a 3D point cloud semantic segmentation algorithm fused with edge detection was proposed. First, the global semantic features in the region are initially extracted from the point cloud data through the 3D point cloud semantic segmentation network. Then, the semantic edge detection network with the introduction of an attention mechanism is adopted, which can better extract and enhance the features of objects in the point cloud data, suppress the generation of non-edge information, and obtain edge features with rich semantic information. Finally, the semantic features belonging to the same object are fused by the fusion module for segmentation and refinement processing, which makes the segmentation target more accurate. In addition, dual semantic loss functions are used, which enables the network to produce semantic segmentation results with better boundaries. By building an experimental platform and using the S3DIS standard data set to test, the average interaction ratio of the improved algorithm on the data set is 70. 21%, which is better than the KPConv semantic segmentation algorithm in precision. The experimental results show that the algorithm can effectively solve the problems of unclear boundary segmentation and fuzzy edge information, and the overall segmentation performance is good. |
| Key words: 3D point cloud semantic segmentation semantic edge detection feature fusion |
