轻量级实时点云分类网络LightPointNet
|
摘要
点云数据的无序性、稀疏性和有限性等特点给基于深度学习的点云模型分类带来了较大的困难.现有的面向点云的深度学习网络存在模型架构复杂、训练参数较多的问题,难以适用于实时点云识别任务,为此提出一种轻量级实时点云网络——LightPointNet.首先,基于点云模型的特点及轻量级点云分类网络的设计原则,提出面向点云模型分类的深度学习网络原型;然后,通过控制变量法完成网络参数设置的优化,形成最终的点云网络LightPointNet.该网络结构紧凑,仅包含3层卷积, 1层池化和1层全连接,且其参数个数不到0.07M.实验结果表明,在ModelNet40上,相比PointNet,VoxNet和LightNet,LightPointNet分类精度分别提高了0.29%,6.49%和2.59%,参数量减少了98.0%,92.4%和76.6%;在MINST和SHREC15上,该网络拥有良好的普适性;这些结果充分证明了LightPointNet分类性能良好且计算效率高,具有轻量级、实时性优点,可以部署在嵌入式设备中,在物联网和点云实时处理等方面具有广阔的应用前景.
The disorder, sparseness and finiteness of point cloud data make it difficult to classify point cloud models based on deep learning. The existing point cloud-oriented deep learning networks have the problems of complex model structures and many training parameters, which make it difficult to apply for real-time point cloud recognition tasks. To address these problems, a lightweight real-time point cloud network, LightPointNet, is proposed. Firstly, based on the characteristics of point cloud models and the design principle of lightweight point cloud classification network, a prototype of deep learning network for point cloud model classification is proposed. Then, the network parameters are optimized and the final point cloud network LightPointNet is formed using variable-controlling approach. The network is compact in structure, consisting of only 3 layers of convolution, 1 layer of pooling and 1 layer of full connection, and the number of parameters is less than 0.07 M. Experiments on ModelNet40 dataset have shown that LightPointNet improve the classification accuracy rates of PointNet, VoxNet, and LightNet by 0.29%, 6.49%, and 2.59%, and its parameter size is reduced by 98.00%, 92.40%,and 76.60%, respectively. Experiments on MINST and SHREC15 have shown that LightPointNet has universal adaptability for wide variety of point cloud data. This result demonstrates that the LightPointNet achieves high classification performance, high computational efficiency, lightweight and real-time advantages. Therefore, the network can be deployed in embedded devices and has a broad application prospect in the Internet of Things,point cloud real-time processing and so on.
The disorder, sparseness and finiteness of point cloud data make it difficult to classify point cloud models based on deep learning. The existing point cloud-oriented deep learning networks have the problems of complex model structures and many training parameters, which make it difficult to apply for real-time point cloud recognition tasks. To address these problems, a lightweight real-time point cloud network, LightPointNet, is proposed. Firstly, based on the characteristics of point cloud models and the design principle of lightweight point cloud classification network, a prototype of deep learning network for point cloud model classification is proposed. Then, the network parameters are optimized and the final point cloud network LightPointNet is formed using variable-controlling approach. The network is compact in structure, consisting of only 3 layers of convolution, 1 layer of pooling and 1 layer of full connection, and the number of parameters is less than 0.07 M. Experiments on ModelNet40 dataset have shown that LightPointNet improve the classification accuracy rates of PointNet, VoxNet, and LightNet by 0.29%, 6.49%, and 2.59%, and its parameter size is reduced by 98.00%, 92.40%,and 76.60%, respectively. Experiments on MINST and SHREC15 have shown that LightPointNet has universal adaptability for wide variety of point cloud data. This result demonstrates that the LightPointNet achieves high classification performance, high computational efficiency, lightweight and real-time advantages. Therefore, the network can be deployed in embedded devices and has a broad application prospect in the Internet of Things,point cloud real-time processing and so on.
引文
[1]Zhi S F,Liu Y X,Li X,et al.Towards real-time 3D object recognition:a lightweight volumetric CNN framework using multitask learning[J].Computers&Graphics,2018,71:199-207
[2]Qi Charles R,Su H,Kaichun M,et al.PointNet:deep learning on point sets for 3D classification and segmentation[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.Los Alamitos:IEEE Computer Society Press,2017:77-85
[3]Maturana D,Scherer S.Voxnet:a 3D convolutional neural network for real-time object recognition[C]//Proceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems.Los Alamitos:IEEE Computer Society Press,2015:922-928
[4]Shilane P,Min P,Kazhdan M,et al.The Princeton shape benchmark[C]//Proceedings of the Shape Modeling Applications.Los Alamitos:IEEE Computer Society Press,2004:167-178
[5]Qi C R,Yi L,Su H,et al.Pointnet++:deep hierarchical feature learning on point sets in a metric space[OL].[2018-06-15].https://arxiv.org/abs/1706.02413
[6]Li Y Y,Bu R,Sun M C,et al.PointCNN:convolution on X-transformed points[OL].[2018-06-15].https://arxiv.org/abs/1801.07791
[7]Vinyals O,Bengio S,Kudlur M.Order matters:sequence to sequence for sets[OL].[2018-06-15].https://arxiv.org/abs/1511.06391
[8]Simonyan K,Zisserman A.Very deep convolutional networks for large-scale image recognition[OL].[2018-06-15].https://arxiv.org/abs/1409.1556
[9]Rusu R B,Cousins S.3D is here:point cloud library(PCL)[C]//Proceedings of the IEEE International Conference on Robotics and Automation.Los Alamitos:IEEE Computer Society Press,2011:1-4
[10]Shi B G,Bai S,Zhou Z C,et al.DeepPano:deep panoramic representation for 3-D shape recognition[J].IEEE Signal Processing Letters,2015,22(12):2339-2343
[11]Su H,Maji S,Kalogerakis E,et al.Multi-view convolutional neural networks for 3D shape recognition[C]//Proceedings of the IEEE International Conference on Computer Vision.Los Alamitos:IEEE Computer Society Press,2015:945-953
[12]Johns E,Leutenegger S,Davison A J.Pairwise decomposition of image sequences for active multi-view recognition[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.Los Alamitos:IEEE Computer Society Press,2016:3813-3822
[13]Wu Z R,Song S R,Khosla A,et al.3D shapenets:a deep representation for volumetric shapes[C]//Proceedings of the IEEEConference on Computer Vision and Pattern Recognition.Los Alamitos:IEEE Computer Society Press,2015:1912-1920
[14]Garcia-Garcia A,Gomez-Donoso F,Garcia-Rodriguez J,et al.PointNet:a 3D convolutional neural network for real-time object class recognition[C]//Proceedings of the International Joint Conference on Neural Networks.Los Alamitos:IEEE Computer Society Press,2016:1578-1584
[15]Sedaghat N,Zolfaghari M,Amiri E,et al.Orientation-boosted voxel nets for 3D object recognition[OL].[2018-06-15].https://arxiv.org/abs/1604.03351
[16]Brock A,Lim T,Ritchie J M,et al.Generative and discriminative voxel modeling with convolutional neural networks[OL].[2018-06-15].https://arxiv.org/abs/1608.04236
[17]Hegde V,Zadeh R.FusionNet:3D object classification using multiple data representations[OL].[2018-06-15].https://arxiv.org/abs/1607.05695
[2]Qi Charles R,Su H,Kaichun M,et al.PointNet:deep learning on point sets for 3D classification and segmentation[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.Los Alamitos:IEEE Computer Society Press,2017:77-85
[3]Maturana D,Scherer S.Voxnet:a 3D convolutional neural network for real-time object recognition[C]//Proceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems.Los Alamitos:IEEE Computer Society Press,2015:922-928
[4]Shilane P,Min P,Kazhdan M,et al.The Princeton shape benchmark[C]//Proceedings of the Shape Modeling Applications.Los Alamitos:IEEE Computer Society Press,2004:167-178
[5]Qi C R,Yi L,Su H,et al.Pointnet++:deep hierarchical feature learning on point sets in a metric space[OL].[2018-06-15].https://arxiv.org/abs/1706.02413
[6]Li Y Y,Bu R,Sun M C,et al.PointCNN:convolution on X-transformed points[OL].[2018-06-15].https://arxiv.org/abs/1801.07791
[7]Vinyals O,Bengio S,Kudlur M.Order matters:sequence to sequence for sets[OL].[2018-06-15].https://arxiv.org/abs/1511.06391
[8]Simonyan K,Zisserman A.Very deep convolutional networks for large-scale image recognition[OL].[2018-06-15].https://arxiv.org/abs/1409.1556
[9]Rusu R B,Cousins S.3D is here:point cloud library(PCL)[C]//Proceedings of the IEEE International Conference on Robotics and Automation.Los Alamitos:IEEE Computer Society Press,2011:1-4
[10]Shi B G,Bai S,Zhou Z C,et al.DeepPano:deep panoramic representation for 3-D shape recognition[J].IEEE Signal Processing Letters,2015,22(12):2339-2343
[11]Su H,Maji S,Kalogerakis E,et al.Multi-view convolutional neural networks for 3D shape recognition[C]//Proceedings of the IEEE International Conference on Computer Vision.Los Alamitos:IEEE Computer Society Press,2015:945-953
[12]Johns E,Leutenegger S,Davison A J.Pairwise decomposition of image sequences for active multi-view recognition[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.Los Alamitos:IEEE Computer Society Press,2016:3813-3822
[13]Wu Z R,Song S R,Khosla A,et al.3D shapenets:a deep representation for volumetric shapes[C]//Proceedings of the IEEEConference on Computer Vision and Pattern Recognition.Los Alamitos:IEEE Computer Society Press,2015:1912-1920
[14]Garcia-Garcia A,Gomez-Donoso F,Garcia-Rodriguez J,et al.PointNet:a 3D convolutional neural network for real-time object class recognition[C]//Proceedings of the International Joint Conference on Neural Networks.Los Alamitos:IEEE Computer Society Press,2016:1578-1584
[15]Sedaghat N,Zolfaghari M,Amiri E,et al.Orientation-boosted voxel nets for 3D object recognition[OL].[2018-06-15].https://arxiv.org/abs/1604.03351
[16]Brock A,Lim T,Ritchie J M,et al.Generative and discriminative voxel modeling with convolutional neural networks[OL].[2018-06-15].https://arxiv.org/abs/1608.04236
[17]Hegde V,Zadeh R.FusionNet:3D object classification using multiple data representations[OL].[2018-06-15].https://arxiv.org/abs/1607.05695