Automated Point Cloud Object Classification

点击翻译按钮获取中文摘要

Delays and cost over-runs in large capital construction projects are often caused by the dimensionalincompatibility of pre-fabricated components such as pipes and couplings. The verification of thesecomponents' dimensions is recognized as a critical process to ensure smooth, on-budget project execution andto eliminate the need for rework. Components must be compared to the original design to ensure as-builtcompliance and virtually matched with the construction site to ensure proper fit upon delivery. Terrestrial laserscanning instruments provide a means to rapidly capture accurate, three-dimensional (3D) datasets (pointclouds) of pre-fabricated components. Current practice for transforming point clouds into 3D models requiresmanual intervention by experienced users. This project aims to automate the process of producing 3D modelsfrom point cloud data. Geometric primitives such as disks and cylinders will be recognized from the pointcloud data contaminated by data clutter, occlusions and noise. The efficient implementation of the proposedsoftware system will benefit the supporting organization and the Canadian construction industry as a whole in anumber of ways. Providing an automatic means for checking dimensional compliance will lead to reduced cutand weld rework will be reduced, more effective use of project manager time, higher quality inspections andminimized construction claims and disputes.