基于亲疏度矩阵的点云置换不变特征提取方法

doi:10.19678/j.issn.1000-3428.0062801

摘要/Abstract

摘要： 点云识别与分割在应用过程中通常需要提取空间旋转不变和置换不变的点云特征，PointCNN采用监督学习的方式来提取，但会产生额外的计算量，PointNet使用最大池化算子提取置换不变特征，且容易忽略点云的局部信息，导致识别准确率下降。提出一种基于亲疏度矩阵的新方法，将点云空间坐标映射到曲率特征空间，提取空间旋转不变特征。通过K邻域内的点集合构建基于欧式内积的亲疏度矩阵，对亲疏度矩阵特征值进行排序，并将对应的特征向量组成变换矩阵，从而变换点云特征，进行置换不变的点云特征卷积操作。在ModelNet40数据集上的实验结果表明，该方法的总体准确率和类别平均准确率分别为92.28%和88.80%，超过PointCNN等基于卷积的方法。此外，该方法通过代数计算的方式获得变换矩阵，可以提高模型训练的效率，且浮点运算数仅为36.6×10⁶ frame/s，大幅降低了训练的复杂度。

关键词: 点云, 三维物体, 旋转不变性, 置换不变性, 神经网络

Abstract: The applicationofpoint cloud recognition and segmentation requires the extraction ofthe spatial rotation invariant and permutation invariant features of the point cloud.PointCNN extracts these features by supervised learning, but this requires additional computation.PointNet extracts permutation-invariant features througha max-pooling operator, but easily ignores the local information of the point cloud, which decreases the recognition accuracy.A new method based on an affinity matrix was proposed to extract spatial rotation-invariant features by mapping the spatial coordinates of the point cloud to the curvature feature space.An affinity matrix based on the Euclidean inner product is built using a point set in K neighborhood.Atransformation matrix is constructed usingthe eigenvectors corresponding to the sorted eigenvalues of the affinity matrix to realize permutation-invariant convolution forfeaturesofthe pointcloud.Experiments ontheMoedlNet40 dataset show that the overall accuracy and mean per-class accuracy of this method are 92.28% and 88.80%, respectively, exceeding othermethods based on convolution, such as thePointCNN.In addition, this method obtains a transformation matrix through algebraic operations, which improves the efficiency of model training.The number of floating-point operations is only 36.6×10⁶ frame/s, which reduces the complexity of training.

Key words: point cloud, three-dimensional object, rotation invariance, permutation invariance, neural networks

中图分类号:

TP391.41

许嘉麟, 姚双, 张蕊华, 许浩, 沈洋. 基于亲疏度矩阵的点云置换不变特征提取方法[J]. 计算机工程, 2022, 48(6): 243-250.

XU Jialin, YAO Shuang, ZHANG Ruihua, XU Hao, SHEN Yang. Permutation Invariant Feature Extraction Method Based on Affinity Matrix of Point Cloud[J]. Computer Engineering, 2022, 48(6): 243-250.

http://www.ecice06.com/CN/Y2022/V48/I6/243

图/表 13

20220625180156

20220625180200

20220625180203

20220625180207

20220625180210

20220625180214

20220625180218

20220625180221

20220625180225

20220625180228

20220625180232

20220625180236

20220625180240

参考文献

[1] SU H, MAJI S, KALOGERAKIS E, et al.Multi-view convolutional neural networks for 3D shape recognition[C]//Proceedings of IEEE International Conference on Computer Vision.Washington D.C., USA:IEEE Press, 2015:945-953.
[2] YANG Z, WANG L W.Learning relationships for multi-view 3D object recognition[C]//Proceedings of IEEE/CVF International Conference on Computer Vision.Washington D.C., USA:IEEE Press, 2019:7504-7513.
[3] 顾砾, 季怡, 刘纯平.基于多模态特征融合的三维点云分类方法[J].计算机工程, 2021, 47(2):279-284. GU L, JI Y, LIU C P.Classification method of three-dimensional point cloud based on multiple modal feature fusion[J].Computer Engineering, 2021, 47(2):279-284.(in Chinese)
[4] MATURANA D, SCHERER S.VoxNet:a 3D convolutional neural network for real-time object recognition[C]//Proceedings of IEEE/RSJ International Conference on Intelligent Robots and Systems.Washington D.C., USA:IEEE Press, 2015:922-928.
[5] WU Z R, SONG S R, KHOSLA A, et al.3D ShapeNets:a deep representation for volumetric shapes[C]//Proceedings of IEEE Conference on Computer Vision and Pattern Recognition.Washington D.C., USA:IEEE Press, 2015:1912-1920.
[6] CHARLES R Q, HAO S, MO K C, et al.PointNet:deep learning on point sets for 3D classification and segmentation[C]//Proceedings of IEEE Conference on Computer Vision and Pattern Recognition.Washington D.C., USA:IEEE Press, 2017:77-85.
[7] QI C R, YI L, SU H, et al.PointNet++:deep hierarchical feature learning on point sets in a metric space[EB/OL].[2021-08-16].https://arxiv.org/abs/1706.02413.
[8] ZAHEER M, KOTTUR S, RAVANBAKHSH S, et al.Deep sets[EB/OL].[2021-08-16].https://arxiv.org/abs/1703.06114.
[9] ESTEVES C, ALLEN-BLANCHETTE C, MAKADIA A, et al.Learning SO(3) equivariant representations with spherical CNNs[J].International Journal of Computer Vision, 2020, 128(3):588-600.
[10] COHEN T S, GEIGER M, KOEHLER J, et al.Spherical CNNs[EB/OL].[2021-08-16].https://arxiv.org/abs/1801.10130.
[11] LI Y, BU R, SUN M, et al.PointcCNN:convolution on X-transformed points[EB/OL].[2021-08-16].https://arxiv.org/abs/1801.07791.
[12] TE G S, HU W, ZHENG A M, et al.RGCNN:regularized graph CNN for point cloud segmentation[C]//Proceedings of the 26th International Conference on Multimedia.New York, USA:ACM Press, 2018:746-754.
[13] ZHANG Y X, RABBAT M.A graph-CNN for 3D point cloud classification[C]//Proceedings of IEEE International Conference on Acoustics, Speech and Signal Processing.Washington D.C., USA:IEEE Press, 2018:6279-6283.
[14] LI J X, CHEN B M, LEE G H.SO-net:self-organizing network for point cloud analysis[C]//Proceedings of IEEE/CVF Conference on Computer Vision and Pattern Recognition.Washington D.C., USA:IEEE Press, 2018:9397-9406.
[15] ZENG W, GEVERS T.3D ContextNet:k-d tree guided hierarchical learning of point clouds using local and global contextual cues[C]//Proceedings of Conference on Computer Vision.Berlin, Germany:Springer, 2019:314-330.
[16] CHEN W K, HAN X G, LI G B, et al.Deep RBFNet:point cloud feature learning using radial basis functions[EB/OL].[2021-08-16].Berlin, Germany:Springer, https://arxiv.org/abs/1812.04302.
[17] ZHAO Y H, BIRDAL T, DENG H W, et al.3D point capsule networks[C]//Proceedings of IEEE/CVF Conference on Computer Vision and Pattern Recognition.Washington D.C., USA:IEEE Press, 2019:1009-1018.
[18] 田钰杰, 管有庆, 龚锐.一种鲁棒的多特征点云分类分割深度神经网络[J].计算机工程, 2021, 47(11):234-240. TIAN Y J, GUAN Y Q, GONG R.A robust deep neural network for multi-feature point cloud classification and segmentation[J].Computer Engineering, 2021, 47(11):234-240.(in Chinese)
[19] SUN X, LIAN Z H, XIAO J G.SRINet:learning strictly rotation-invariant representations for point cloud classification and segmentation[C]//Proceedings of the 27th ACM International Conference on Multimedia.New York, USA:ACM Press, 2019:980-988.
[20] THOMAS N, SMIDT T, KEARNES S, et al.Tensor field networks:Rotationand translation equivariant neural networks for 3D point clouds[EB/OL].[2021-08-16].https://arxiv.org/abs/1802.08219.
[21] POULENARD A, RAKOTOSAONA M J, PONTY Y, et al.Effective rotation-invariant point CNN with spherical harmonics kernels[C]//Proceedings of International Conference on 3D Vision.Washington D.C., USA:IEEE Press, IEEE Press, 2019:47-56.
[22] BOULCH A.ConvPoint:continuous convolutions for point cloud processing[J].Computers & Graphics, 2020, 88:24-34.
[23] LIU Y C, FAN B, XIANG S M, et al.Relation-shape convolutional neural network for point cloud analysis[C]//Proceedings of IEEE/CVF Conference on Computer Vision and Pattern Recognition.Washington D.C., USA:IEEE Press, 2019:8887-8896.
[24] WU W X, QI Z A, LI F X.PointConv:deep convolutional networks on 3D point clouds[C]//Proceedings of IEEE/CVF Conference on Computer Vision and Pattern Recognition.Washington D.C., USA:IEEE Press, 2019:9613-9622.
[25] WU P X, CHEN C, YI J R, et al.Point cloud processing via recurrent set encoding[C]//Proceedings of AAAI Conference on Artificial Intelligence.New York, USA:AAAI Press, 2019:5441-5449.
[26] LIN M, CHEN Q, YAN S.Network in network[EB/OL].[2021-08-16].https://arxiv.org/abs/1312.4400.
[27] LECUN Y, BOTTOU L, BENGIO Y, et al.Gradient-based learning applied to document recognition[J].Proceedings of the IEEE, 1998, 86(11):2278-2324.
[28] WU S H, HUANG H, GONG M L, et al.Deep points consolidation[J].ACM Transactions on Graphics, 2015, 34(6):1-13.
[29] HE T, HUANG H B, YI L, et al.GeoNet:deep geodesic networks for point cloud analysis[C]//Proceedings of 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition.Washington D.C., USA:IEEE Press, 2019:6888-6897.

选择文件类型/文献管理软件名称

选择包含的内容