摘要
针对室内稠密点云数据海量、信息冗余、处理难度大等问题,提出一种顾及平面特征的室内稠密点云精简方法。首先,通过无效点去除、统计滤波、体素滤波等完成稠密点云的格式检查、去噪和抽稀。然后,采用区域生长算法提取平面,并采用Alpha Shape算法提取其边缘信息。其次,将点云数据降维成图像,并提出一种融合聚类的四叉树分割方法实现目标聚类分割。最后,针对图像中不同聚类区域,采用3种采样策略回溯得到精简点云。试验选取公寓、卧室、会议室和办公室等典型室内场景测试方法性能。结果表明:与传统的随机采样、距离采样、八叉树采样等方法相比,该方法简化效果更佳,平均简化误差在3 mm以内。在保留场景平面特征和边缘细节信息的同时,显著降低点云存储空间。对于三维重建、地图管理和机器人导航有着重要意义。
In order to solve the problems of the large storage, information redundancy, and difficult processing of dense point clouds, this paper proposes an indoor dense point cloud simplification method considering plane features.First, invalid point removal, statistical filtering, and voxel filtering are applied to accomplish the format check, denoising, and thinning for the dense point cloud.Then, the region-growing algorithm is used to extract planes and the Alpha Shape to segment its edge information.And, the planar point cloud is reduced into images, and a quadtree segmentation method based on cluster fusion is proposed to achieve target clustering segmentation.Finally, three different resampling strategies among image clusters are adopted to backtrack the simplified point cloud.Typical indoor datasets involving an apartment, a bedroom, a boardroom, and an office were used in the experiments to verify the proposed method.Experimental results showed that an average simplified error was limited within 3 mm, which outperformed the results by random sampling, distance sampling, and octree sampling.The proposed method significantly reduced the storage memory of the point cloud while preserving the planar features and edge details in the scenario.It is significant for the subsequent 3-dimension reconstruction, map management, and robot navigation.
作者
王莉敏
史鹏程
WANG Limin;SHI Pengcheng(Geo-information Center of Henan Province,Zhengzhou 450003,China;School of Computer Science,Wuhan University,Wuhan 430072,China)
出处
《西安科技大学学报》
CAS
北大核心
2023年第1期183-191,共9页
Journal of Xi’an University of Science and Technology
基金
辽宁省教育厅服务地方项目(LJ2019FL008)。
关键词
稠密点云
点云精简
区域生长
边缘提取
四叉树分割
dense point cloud
point cloud simplification
region growing
boundary extraction
quadtree segmentation
