采用动态残差图卷积的3D点云超分辨率

为了在超分辨率中对非欧数据的3D点云进行局部信息高效提取，提出采用动态残差图卷积的3D点云超分辨率网络(PSR-DRGCN)，该网络包括特征提取模块、DRGCN模块及上采样模块. 对于输入的3D点云, 特征提取模块采用k-近邻(k-NN)算法在3D空间中找到每个点对应的k个邻居，通过逐点卷积把局部几何信息转换到高维特征空间中；DRGCN模块利用多层图卷积操作将3D空间中每个点的局部几何特征抽象为语义特征，在每一层对点的近邻空间进行动态调整以增加感受野范围，并通过残差连接融合多层次语义信息，从而对局部几何信息高效提取；上采样模块将特征空间中的点进行上采样并转换到3D空间中. 实验结果表明，PSR-DRGCN生成的高分辨率点云在放大尺度为2倍时，相似性指标CD、EMD、F-score相比第2网络分别优化了10.00%，4.76%，16.84%；当放大尺度为6倍时，相似性指标相比第2网络分别优化了2.35%，40.00%，0.58%；在所有情况下的均值与标准差指标上达到最优效果，生成的高分辨率点云质量高.

3D point cloud super-resolution with dynamic residual graph convolutional networks

A 3D point cloud super-resolution network with dynamic residual graph convolution (PSR-DRGCN) was proposed to efficiently extract of local information from 3D point clouds of non-European data in super-resolution. The network includes feature extraction module, DRGCN module and upsampling module. For the input point cloud, the feature extraction module locates k nearest points of each point in 3D space by k-NN algorithm and then converts the local geometry information into the high dimensional feature space through a multi-layer pointwise convolution. The DRGCN module converts the local geometry feature of each point into the semantic feature through a multi-layer graph convolution. It dynamically adjusts the neighbor space of the point in each layer to increase the receptive field range and effectively fuse the semantic information of different levels through residual connection, which makes the extraction of local geometric information efficient. The upsampling module adds the number of points and maps them from feature space to 3D space. The results showed that at 2× magnification of the high-resolution point cloud generated by PSR-DRGCN, the similarity indexes CD, EMD and F-score compared with the second network were increased by 10.00%, 4.76% and 16.84% respectively. Compared with the second network, the similarity indexes at 6× magnification were increased by 2.35%, 40.00% and 0.58% respectively. In all cases, the optimal effect was achieved on the mean and the std indicators and the generated high-resolution point cloud quality was high.