| 387 | 4 | 160 |
| 下载次数 | 被引频次 | 阅读次数 |
激光雷达采集的点云数据往往是稀疏且无序的。对于需要通过处理三维点云的研究来说，直接处理激光雷达采集的数据容易产生差错，需要对采集的点云进行预处理。现有算法致力于恢复点云的拓扑结构，忽略了稀疏的点云容易丢失特征信息。针对上述问题，提出了一个神经网络，可以将原来残缺、稀疏的点云生成为完整、密集、均匀的点云，称为点云补全均匀化网络(Points Completion Uniform Net, PCU-Net)。该网络基于点云补全网络(Point Completion Network, PCN)引入了一种能够快速提取全局特征的轻量化结构，并在解码器中补全和稠密化点云。还提出了一种精炼器模块，从输入中保留原始细节，通过最远点采样(Farthest Point Sampling, FPS)和点特征残差网络均匀化点云。在开源数据集Visionair上通过实验对比，该算法在点云补全上较目前主流补全算法有所提升，并在点云均匀化和稠密化上取得良好的效果。
Abstract：The point cloud data collected by lidar is often sparse and disordered. For the research that needs to process three-dimensional point clouds, it is to make mistakes by directly processing the data collected by lidar, so it is necessary to preprocess the collected point clouds. The existing algorithms focus on restoring the topology of point clouds, while ignoring that sparse point clouds are prone to lose feature information. To solve the above problems, a neural network is proposed, which can generate a complete, dense and uniform point cloud which called Points Completion Uniform Net(PCU-Net) by using the original incomplete and sparse point cloud. In this network, based on Point Completion Network(PCN),a lightweight structure is introduced to extract global features quickly and complete and densify the point cloud in the decoder. In addition, a refiner module is proposed, which preserves the original details from the input and unifies the point cloud through the farthest point sampling and the point feature residual network. Through experimental comparison on Visionair open source data set, the point cloud completion is improved by the proposed algorithm compared with the current mainstream completion algorithm, and the proposed algorithm can achieve good results in point cloud homogenization and densification.
[1] 范小辉，许国良，李万林，等.基于深度图的三维激光雷达点云目标分割方法[J].中国激光，2019,46(7):292-299.
[2] 卢宏涛，张秦川.深度卷积神经网络在计算机视觉中的应用研究综述[J].数据采集与处理，2016,31(1):1-17.
[3] CHARLES Q C R,SU H,MO K,et al.PointNet:Deep Learning on Point Sets for 3D Classification and Segmentation[C]∥IEEE Conference on Computer Vision and Pattern Recognition.Honolulu:IEEE,2017:77-85.
[4] YUAN W T,KHOT T,HELD D,et al.PCN:Point Completion Network[C]//2018 International Conference on 3D Vision (3DV).Verona:IEEE,2018:728-737.
[5] TCHAPMI L P,KOSARAJU V,REZATOFIGHI H,et al.Topnet:Structural Point Cloud Decoder[C]//IEEE/CVF Conference on Computer Vision and Pattern Recognition.Long Beach:IEEE/CVF,2019:383-392.
[6] HUANG Z T,YU Y K,XU J W,et al.PF-Net:Point Fractal Network for 3D Point Cloud Completion[C]//IEEE/CVF Conference on Computer Vision and Pattern Recognition.Seattle:IEEE/CVF,2020:7659-7667.
[7] ZHANG Q L,YANG Y B.SA-Net:Shuffle Attention for Deep Convolutional Neural Networks[C]∥ICASSP 2021-2021 IEEE International Conference on Acoustics,Speech and Signal Processing (ICASSP).Toronto:IEEE,2021:2235-2239.
[8] WEN X,XIANG P,HAN Z Z,et al.PMP-Net:Point Cloud Completion by Learning Multi-step Point Moving Paths[C]∥2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition.Nashville:IEEE/CVF,2021:7443-7452.
[9] YU L Q,LI X Z,FU C W,et al.PU-Net:Point Cloud Upsampling Network[C]∥IEEE/CVF Conference on Computer Vision and Pattern Recognition.Salt Lake City:IEEE/CVF,2018:2790-2799.
[10] LI R H,LI X Z,FU C W,et al.PU-GAN:A Point Cloud Upsampling Adversarial Network[C]//2019 IEEE/CVF International Conference on Computer Vision.Seoul:IEEE/CVF:2019:7202-7211.
[11] CHARLES Q C R,YI L,SU H,et al.PointNet++:Deep Hierarchical Feature Learning on Point Sets in a Metric Space[J/OL].(2017-06-07)[2022-10-30].https:∥arxiv.org/abs/1706.02413.
[12] HU Q Y,YANG B,XIE L H,et al.RandLA-Net:Efficient Semantic Segmentation of Large-scale Point Clouds[C]//IEEE/CVF Conference on Computer Vision and Pattern Recognition.Seattle:IEEE/CVF,2020:11105-11114.
[13] PETERSON L E.K-nearest Neighbor[J].Scholarpedia,2009,4(2):1883.
[14] ZHANG C,PAN X,LI H P,et al.A Hybrid MLP-CNN Classifier for Very Fine Resolution Remotely Sensed Image Classification[J].ISPRS Journal of Photogrammetry and Remote Sensing,2018,140:133-144.
[15] LIANG X Z,WANG X B,LEI Z,et al.Soft-margin Softmax for Deep Classification[C]// International Conference on Neural Information Processing.Guangzhou:Springer,2017:413-421.
[16] LIU M H,SHENG L,YANG S,et al.Morphing and Sampling Network for Dense Point Cloud Completion[C]//Proceedings of the AAAI Conference on Artificial Intelligence.New York:AAAI,2020:11596-11603.
[17] YANG Y,FENG C,SHEN Y,et al.FoldingNet:Point Cloud Auto-encoder via Deep Grid Deformation[C]//2018 IEEE Conference on Computer Vision and Pattern Recognition.Salt Lake City:IEEE,2018:206-215.
[18] 赵新灿，常寒星，金仁标.3D点云形状补全GAN[J].计算机科学，2021,48(4):192-196.
[19] LI S H,JIAO J T,HAN Y J,et al.Demystifying ResNet[J/OL].(2017-05-20)[2022-10-30].https://arxiv.org/abs/1611.01186v2.
[20] WANG K F,GOU C,DUAN Y J,et al.Generative Adversarial Networks:Introduction and Outlook[J].IEEE/CAA Journal of Automatica Sinica,2017,4(4):588-598.
[21] WU T,PAN L,ZHANG J Z,et al.Balanced Chamfer Distance as a Comprehensive Metric for Point Cloud Completion[J/OL].[2022-10-30].https://www.xueshufan.com/publication/3214314028.
[22] LIN T Y,DOLLáR P,GIRSHICK R,et al.Feature Pyramid Networks for Object Detection[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition.Honolulu:IEEE,2017:2117-2125.
基本信息:
DOI：
中图分类号:TN958.98
引用信息:
[1]郎超豪,甘兴利,施浩.PCU-Net:基于改进PCN的点云补全均匀算法[J].无线电工程,2023,53(06):1269-1274.
基金信息:
国家自然科学基金(62101088)~~