Research on Point Cloud Segmentation of Ship Segment Closure Surface Based on PointSIFT
陈尚伟① CHEN Shang-wei；汪骥①WANG Ji；史卫东② SHI Wei-dong；
陈奎② CHEN Kui；刘兵兵② LIU Bing-bing
（①大连理工大学船舶工程学院，大连 116024；②大连船舶重工集团有限公司，大连 116021）
（①School of Naval Architecture & Ocean Engineering，Dalian University of Technology，Dalian 116024，China；②Dalian Shipbuilding Industry Co.，Ltd.，Dalian 116021，China）
摘要：船体分段合拢面的精度检测十分重要。三维扫描仪在进行分段合拢面的扫描时，同样会冗余地把周围分段甚至环境的点记录下来。基于此，文章从辅助三维扫描仪进行精度检测的角度出发，对于三维扫描仪扫描出的合拢面点云进行分割，对点云分割网络PointSIFT 进行了适当的改进，从 CAD 中导出点云数据并开发软件完成标注工作，采用随机梯度下降算法完成了对网络的训练。实验证明，网络模型对于验证集的精确率为 76%，召回率为 92%，点云分割网络同样适用于船体分段这种大尺寸物体的分割。
Abstract: The accuracy detection of the hull section closure surface is very important. When scanning segmented surfaces, 3D scanners also redundantly record points of surrounding segments and even the environment. Based on this, the article starts from the perspective of assisting 3D scanners in accuracy detection, segments the closed surface point cloud scanned by the 3D scanner, and makes appropriate improvements to the point cloud segmentation network PointSIFT. Point cloud data is exported from CAD and software is developed to complete annotation work. Random gradient descent algorithm is used to train the network. Experimental results have shown that the network model has an accuracy rate of 76% and a recall rate of 92% for the validation set. The point cloud segmentation network is also suitable for segmenting large-sized objects such as ship hull segmentation.
关键词：点云；PointSIFT；船体分段合拢面；精度检测
Key words: point cloud；PointSIFT；sectional closure surface of the hull；accuracy testing
中图分类号：U671.99 文献标识码：A 文章编号：1006-4311（2024）30-131-04 doi:10.3969/j.issn.1006-4311.2024.30.038
|
