基于先验知识的三维点云鲁棒处理技术研究

项目名称： 基于先验知识的三维点云鲁棒处理技术研究

项目编号： No.61202334

项目类型： 青年科学基金项目

立项/批准年度： 2013

项目学科： 计算机科学学科

项目作者： 李宝

作者单位： 中国人民解放军国防科学技术大学

项目金额： 23万元

中文摘要： 鲁棒的三维点云处理技术是促进三维模型获取技术在多个领域广泛应用的关键因素之一。然而，实际应用中各种获取技术得到的点云不可避免地包含噪声、外点、数据缺失以及不均匀采样，这些缺陷给现有的点云处理方法带来了诸多挑战。为此，本项目拟以构造点云的结构化表示为目标，借助人类关于物体形状和结构的先验知识，基于三维重建、高层次形状分析等领域的最新研究成果，通过引入对称性分析、重复模式检测等技术有效地消除噪声、去除外点、修复缺失并改善分布密度，并借助法向量信息实现特征保持的点云增强，然后提取反映点云几何特征和拓扑特征的线性特征，最后利用基于RANSAC的形状基元检测技术对点云进行拟合并构建基元之间的关系。本研究可有效解决当前点云处理方法鲁棒性不足的问题，为三维模型的重建、编辑以及三维物体的识别与理解提供良好输入，从而促进三维模型获取技术的应用与普及。

中文关键词： 三维点云；先验知识；鲁棒处理；特征提取；结构化表示

英文摘要： Robust processing of 3D point clouds is an essential factor which renders the applying of 3D model acquisition methods in various applications of many fields. However, the point clouds obtained via different acquisition methods usually inevitably contain noise, outliers, holes and uneven distribution. These defects bring many challenges to existing point clouds processing techniques. Therefore, aiming at building structural representations of point clouds, this project propose to resort to human prior knowledge about shape and structures of objects in the real world, as well as the state-of-the-art work of 3D model reconstruction and shape analysis, to investigate and propose robust processing techniques of point clouds. With the help of symmetry analysis and repetition detection methods, the proposed method can achieve denoisng, outlier removal, hole-filling and better distribution density. Meanwhile, normal information is used to achieve feature preserving on the consolidated point clouds. Then linear features which can capture the geometry or topology of point clouds are extracted. Finally, RANSAC based shape primitive detection is used to fit the point clouds and the relations of primitives are recovered simultaneously. This project can achieve higher robustness which is an overall limitation of existing met

英文关键词： 3D Point Cloud；Prior Knowledge；Robust Processing；Feature Extraction；Structural Representation