基于层级边缘卷积的三维点云分类

3D point cloud classification based on hierarchical edge convolution

针对现有算法存在局部信息提取不充分、区域信息合并有限的问题,提出了一种分层捕获局部几何特征的点云分类算法,对点云进行下采样,构建局部区域,对局部区域中的点与点之间的特征距离进行建模,获得几何信息,并利用层级结构对特征进行有效归纳,预测点云模型所属类别。理论分析和实验结果表明,与已有的点云分类算法相比,所提算法能够更充分地提取局部信息,在提高点云模型分类准确率的同时实现复杂度和准确率的平衡。

To solve the problems of insufficient local information extraction and limited integration of regional information in exist algorithms,a point cloud classification algorithm for hierarchically capturing local geometric features was proposed.The algorithm down-sampled point cloud and constructed local area.The geometric information was obtained according feature distances between points in the local area.Then the hierarchical structure was used to effectively summarize the features and predict the category of the point cloud model.Compared with the existing point cloud classification algorithms,the proposed algorithm could extract local information more sufficiently.The experimental results show that this algorithm improves point cloud model classification accuracy and achieves the balance between complexity and accuracy.