一种改进的空间上下文点云分类方法

摘要

To address the lacking of effectively utilization of nonlocal spatial context information on complex scene when classifying point cloud, an improved contextual classification method is proposed for point cloud with linear distribution and uneven density.Firstly, the local point cloud features and interaction spatial context were estimated based on the curvature based adaptive neighborhoods.Then, the supervoxel based distribution spatial context was extracted from point cloud.Finally, the point cloud classification was achieved automatically via higher-order conditional random field, which overcomes the limitation of local feature based point cloud classification.The experimental results show that the proposed method is able to improve the accuracy of point cloud classification effectively.%考虑到点云数据具有线性分布和密度不均匀的特点,以及现有复杂场景点云分类方法中缺少对非局部空间上下文信息的有效利用,提出了一种改进的空间上下文点云分类方法.该方法在提取点云数据顾及曲率的自适应邻域的基础上,首先估算点云局部特征与依赖性空间上下文,并基于超级体素提取分布性空间上下文,最后采用高阶条件随机场模型,实现对点云数据的自动分类,避免了利用单一点云局部特征分类的局限性.试验结果表明,本文方法能够有效提高点云数据地物分类精度.