基于区域分割的多视角点云精简算法-《计算机技术与发展》

文章信息/Info

Title:: Multi-view Point Cloud Reduction Algorithm Based onRegion Segmentation

作者:: 梁正友¹; 2 ; 黄思捷¹ ; 孙宇¹; 李轩昂¹; 1. 广西大学计算机与电子信息学院,广西南宁 530004;
2. 广西多媒体通信与网络技术重点实验室,广西南宁 530004

Author(s):: LIANG Zheng-you¹; 2 ; HUANG Si-jie¹ ; SUN Yu¹ ; LI Xuan-ang¹; 1. School of Computer and Electronic Information,Guangxi University,Nanning 530004,China;
2. Guangxi Key Laboratory of Multimedia Communications and Network Technology,Nanning 530004,China

Keywords:: multi-view point cloud; point cloud simplification; Kinect; group random reduction; bounding box

摘要:: 由于多视角点云的重叠区域点云密度过大,导致出现占用过多的存储空间和降低计算机效率等问题。针对此问题,提出一种基于重叠区域分割和分组随机精简的 Kinect 多视角点云精简算法。首先,用包围盒子提取相邻点云的重叠区域内的点云,将它们分割成重叠区域点云和非重叠区域点云。其次,用一个分组随机精简算法对重叠区域点云进行精简。最后,对所有的重叠区域点云精简之后,与非重叠区域点云一起合并成一个整体点云,根据精简率的要求再对整体点云进行一次精简。用 Kinect 分别采集水果篮子和打印机等两种物体点云,用得到的多视角点云对提出的算法进行了实验验证。实验结果表明,在减少对目标物体点云的细节破坏和保留目标物体的点云细节和特征等方面,该算法优于基于包围盒子的精简算法。

Abstract:: Because the density of point cloud in the overlapping area of multi-view point cloud is too large,it takes up too much storage space and reduces computer efficiency. To solve this problem,a Kinect multi-view point cloud reduction algorithm based on overlapping region segmentation and grouping random reduction is proposed. Firstly,the point clouds in the overlapped area of the adjacent point clouds are extracted by bounding box,which are divided into the overlapped area point clouds and the non-overlapped area point clouds.Secondly,a group random reduction algorithm is used to simplify the point cloud in the overlapping area. Finally,after all overlapped area point clouds are simplified,they are combined with non-overlapped area point clouds to form a whole point cloud. According to the requirements of the simplification rate,the whole point cloud is simplified again. Kinect was used to collect point clouds of fruit basket sand printers, and the obtained multi-view point clouds were used to experimentally verify the proposed algorithm. The experiment shows that the algorithm is superior to the bounding box based algorithm in terms of reducing the damage to the details of the target object and preserving the details and features of the target object.