用于激光雷达语义分割的体素补全和三维非对称卷积网络

IF 3 4区 计算机科学 Q2 COMPUTER SCIENCE, INFORMATION SYSTEMS
Yan Zhou, Jingwei Liu, Jianxun Li, Haibin Zhou
{"title":"用于激光雷达语义分割的体素补全和三维非对称卷积网络","authors":"Yan Zhou, Jingwei Liu, Jianxun Li, Haibin Zhou","doi":"10.1007/s11042-024-19975-2","DOIUrl":null,"url":null,"abstract":"<p>The point cloud data collected by LiDAR is large in scale and contains rich spatial structure detail information, through the collection and labeling of LiDAR data, the automatic driving system can obtain detailed information about the environment around the vehicle. Due to lack of sufficient laser points, some methods transform the point cloud to dense representations such as multi-view or voxelized grids for processing, ignoring the information loss problem caused by the LiDAR imaging characteristics as well as the point cloud transformations, which leads to a degradation of the segmentation performance. In this work, We investigate a 3D semantic segmentation scheme with only LiDAR inputs, called voxel completion and 3D asymmetric convolution network. We propose a voxel completion sub-network to improve the feature extraction capability of the network by enlarging the receptive field and using multi-scale feature extraction to reduce the empty units in the voxels and obtain more complete voxel features. In addition, due to the presence of a large number of cubic objects in the autopilot scenario, to better match the autopilot scenario, we propose a 3D asymmetric convolution network that includes three components: a 3D residual block, an asymmetric convolution block, and a context module. These components are combined together to explore 3D geometric patterns, which can maintain their intrinsic properties and improve the performance of the network. Extensive experiments on the SemanticKITTI and nuScenes benchmark datasets demonstrate the superiority of the approach. For example, on the nuScenes validation set, our method outperforms the state-of-the-art method by 0.3% in mIoU.</p>","PeriodicalId":18770,"journal":{"name":"Multimedia Tools and Applications","volume":null,"pages":null},"PeriodicalIF":3.0000,"publicationDate":"2024-09-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"Voxel completion and 3D asymmetrical convolution networks for Lidar semantic segmentation\",\"authors\":\"Yan Zhou, Jingwei Liu, Jianxun Li, Haibin Zhou\",\"doi\":\"10.1007/s11042-024-19975-2\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"<p>The point cloud data collected by LiDAR is large in scale and contains rich spatial structure detail information, through the collection and labeling of LiDAR data, the automatic driving system can obtain detailed information about the environment around the vehicle. Due to lack of sufficient laser points, some methods transform the point cloud to dense representations such as multi-view or voxelized grids for processing, ignoring the information loss problem caused by the LiDAR imaging characteristics as well as the point cloud transformations, which leads to a degradation of the segmentation performance. In this work, We investigate a 3D semantic segmentation scheme with only LiDAR inputs, called voxel completion and 3D asymmetric convolution network. We propose a voxel completion sub-network to improve the feature extraction capability of the network by enlarging the receptive field and using multi-scale feature extraction to reduce the empty units in the voxels and obtain more complete voxel features. In addition, due to the presence of a large number of cubic objects in the autopilot scenario, to better match the autopilot scenario, we propose a 3D asymmetric convolution network that includes three components: a 3D residual block, an asymmetric convolution block, and a context module. These components are combined together to explore 3D geometric patterns, which can maintain their intrinsic properties and improve the performance of the network. Extensive experiments on the SemanticKITTI and nuScenes benchmark datasets demonstrate the superiority of the approach. For example, on the nuScenes validation set, our method outperforms the state-of-the-art method by 0.3% in mIoU.</p>\",\"PeriodicalId\":18770,\"journal\":{\"name\":\"Multimedia Tools and Applications\",\"volume\":null,\"pages\":null},\"PeriodicalIF\":3.0000,\"publicationDate\":\"2024-09-16\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"Multimedia Tools and Applications\",\"FirstCategoryId\":\"94\",\"ListUrlMain\":\"https://doi.org/10.1007/s11042-024-19975-2\",\"RegionNum\":4,\"RegionCategory\":\"计算机科学\",\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"Q2\",\"JCRName\":\"COMPUTER SCIENCE, INFORMATION SYSTEMS\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"Multimedia Tools and Applications","FirstCategoryId":"94","ListUrlMain":"https://doi.org/10.1007/s11042-024-19975-2","RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q2","JCRName":"COMPUTER SCIENCE, INFORMATION SYSTEMS","Score":null,"Total":0}
引用次数: 0

摘要

激光雷达采集的点云数据尺度大,包含丰富的空间结构细节信息,通过对激光雷达数据的采集和标注,自动驾驶系统可以获得车辆周围环境的详细信息。由于缺乏足够的激光点,一些方法将点云转换为多视角或体素化网格等密集表示形式进行处理,忽略了激光雷达成像特性以及点云转换带来的信息丢失问题,导致分割性能下降。在这项工作中,我们研究了一种仅使用激光雷达输入的三维语义分割方案,称为体素完成和三维非对称卷积网络。我们提出了一个体素完成子网络,通过扩大感受野和使用多尺度特征提取来减少体素中的空单元,获得更完整的体素特征,从而提高网络的特征提取能力。此外,由于自动驾驶场景中存在大量立方体物体,为了更好地匹配自动驾驶场景,我们提出了一种三维非对称卷积网络,其中包括三个组件:三维残差块、非对称卷积块和上下文模块。这些组件结合在一起,共同探索三维几何模式,既能保持其固有特性,又能提高网络性能。在 SemanticKITTI 和 nuScenes 基准数据集上进行的大量实验证明了这种方法的优越性。例如,在 nuScenes 验证集上,我们的方法在 mIoU 方面比最先进的方法高出 0.3%。
本文章由计算机程序翻译,如有差异,请以英文原文为准。

Voxel completion and 3D asymmetrical convolution networks for Lidar semantic segmentation

Voxel completion and 3D asymmetrical convolution networks for Lidar semantic segmentation

The point cloud data collected by LiDAR is large in scale and contains rich spatial structure detail information, through the collection and labeling of LiDAR data, the automatic driving system can obtain detailed information about the environment around the vehicle. Due to lack of sufficient laser points, some methods transform the point cloud to dense representations such as multi-view or voxelized grids for processing, ignoring the information loss problem caused by the LiDAR imaging characteristics as well as the point cloud transformations, which leads to a degradation of the segmentation performance. In this work, We investigate a 3D semantic segmentation scheme with only LiDAR inputs, called voxel completion and 3D asymmetric convolution network. We propose a voxel completion sub-network to improve the feature extraction capability of the network by enlarging the receptive field and using multi-scale feature extraction to reduce the empty units in the voxels and obtain more complete voxel features. In addition, due to the presence of a large number of cubic objects in the autopilot scenario, to better match the autopilot scenario, we propose a 3D asymmetric convolution network that includes three components: a 3D residual block, an asymmetric convolution block, and a context module. These components are combined together to explore 3D geometric patterns, which can maintain their intrinsic properties and improve the performance of the network. Extensive experiments on the SemanticKITTI and nuScenes benchmark datasets demonstrate the superiority of the approach. For example, on the nuScenes validation set, our method outperforms the state-of-the-art method by 0.3% in mIoU.

求助全文
通过发布文献求助,成功后即可免费获取论文全文。 去求助
来源期刊
Multimedia Tools and Applications
Multimedia Tools and Applications 工程技术-工程:电子与电气
CiteScore
7.20
自引率
16.70%
发文量
2439
审稿时长
9.2 months
期刊介绍: Multimedia Tools and Applications publishes original research articles on multimedia development and system support tools as well as case studies of multimedia applications. It also features experimental and survey articles. The journal is intended for academics, practitioners, scientists and engineers who are involved in multimedia system research, design and applications. All papers are peer reviewed. Specific areas of interest include: - Multimedia Tools: - Multimedia Applications: - Prototype multimedia systems and platforms
×
引用
GB/T 7714-2015
复制
MLA
复制
APA
复制
导出至
BibTeX EndNote RefMan NoteFirst NoteExpress
×
提示
您的信息不完整,为了账户安全,请先补充。
现在去补充
×
提示
您因"违规操作"
具体请查看互助需知
我知道了
×
提示
确定
请完成安全验证×
copy
已复制链接
快去分享给好友吧!
我知道了
右上角分享
点击右上角分享
0
联系我们:info@booksci.cn Book学术提供免费学术资源搜索服务,方便国内外学者检索中英文文献。致力于提供最便捷和优质的服务体验。 Copyright © 2023 布克学术 All rights reserved.
京ICP备2023020795号-1
ghs 京公网安备 11010802042870号
Book学术文献互助
Book学术文献互助群
群 号:481959085
Book学术官方微信