三维数据降维处理的实现方法  

Realization method of dimensionality reduction processing of 3d data

作　　者：朱宗玖[1] 裴善强 周之琪 曹亮 ZHU Zong-jiu;PEI Shan-qiang;ZHOU Zhi-qi;CAO Liang(School of Electrical and Information Engineering,Anhui University of Science and Technology,Huainan 232001,China;Beijing Qingying Machine Vision Technology co.,LTD.,Beijing 100044,China)

机构地区：[1]安徽理工大学电气与信息工程学院,安徽淮南232001 [2]北京清影机器视觉技术有限公司,北京100044

出　　处：《光电子．激光》2020年第1期89-95,共7页Journal of Optoelectronics·Laser

摘　　要：针对目前三维点云数据在进行背景分离、数据拼接、尺寸测量和三维重建时均需要人工参与和计算速率特别慢等问题,提出一种三维点云数据降维的处理方法。首先将采集到的三维点云数据建立新的被视空间,确立新的深度、宽度和高度,然后通过坐标的平移转换,分别得到前视投影函数、右视投影函数和俯视投影函数,以及三个方向的深度值函数,最后通过数字化处理,将六个函数转换成二维图片的格式。再通过试验将最终的正向灰度图和三维点云图的数据的占存大小和数据量,误差大小和运算时间作对比,可以得出该降维方法完全继承现有二维图像的各类算法,能快速实现三维数据的各类管理和计算要求;实现了三维数据的简易可视化,便于数据质量的直观检查。Aiming at the problems of background separation,data splicing,dimensional measurement and 3 D reconstruction,the manual 3 D point cloud data needs to be manually involved and the calculation rate is particularly slow.A method for processing dimensional reduction of 3 D point cloud data is proposed.Firstly,the acquired 3 D point cloud data is used to establish a new view space,and a new depth,width and height are established.Then,through the translational transformation of the coordinates,the forward view projection function,the right view projection function and the overhead projection function are respectively obtained,and The depth value function of the direction,and finally the six functions are converted into a two-dimensional picture format by digitization.Then through experiment,the data of the final forward gray scale and the 3 D point cloud map are compared with the data size,the error magnitude and the operation time.It can be concluded that the dimension reduction method completely inherits the existing two-dimensional images.The algorithm can quickly realize various management and calculation requirements of three-dimensional data;realizes simple visualization of three-dimensional data,and facilitates visual inspection of data quality.

关 键 词：三维成像 降维处理 机器视觉 点云数据 图像处理 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]